Good morning, {{ first_name | AI enthusiasts }}. SpaceX’s splashy IPO debut was less than a week ago, but the surging stock has already helped shore up another part of Elon Musk’s empire.

The $SPCX windfall just bankrolled a $60B takeover of Cursor, a deal Musk optioned back in April — and if CEO Michael Truell’s teasers on its next Opus-sized, “generally intelligent” model hold true, Musk may be fast-tracking his way near the frontier.

In today’s AI rundown:

• SpaceX nabs Cursor in $60B all-stock deal

• Z AI’s powerful open-weights model nears the frontier

• Conduct better stock research with Perplexity Finance

• Meta confronts its AI morale crisis

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

SPACEX & CURSOR

💰 SpaceX nabs Cursor in $60B all-stock deal

Image source: Cursor

The Rundown: SpaceX just officially exercised its option to acquire AI coding startup Cursor in a $60B deal paid entirely in stock, coming on the heels of a post-IPO rally that has nearly doubled the company’s value and sent Elon Musk’s personal wealth surging above $1T.

The details:

• Cursor and SpaceX announced an initial deal in April that provided an option to acquire the startup for $60B or pay $10B for the partnership alone.

• Since listing Friday at $135 a share, SpaceX has rocketed past $200, piling nearly $1T onto its value in under a week of public trading.

• Cursor CEO Michael Truell said its upcoming model will be “generally intelligent,” trained from scratch, and as big as Opus.

• SpaceX said Cursor is already part of a model-training push for Grok Build and its own editor, giving Musk a developer platform for his AI stack.

Why it matters: Cursor was already intertwined with SpaceX from the April deal, but the company’s up-only IPO surge just made a stock-only deal even easier. While Grok hasn’t been up to par on coding, the combination of Cursor’s models and SpaceX’s compute could turn things around at the breakneck speeds Musk likes to move at.

TOGETHER WITH MICROSOFT

👀 The FOMO is real: Microsoft Build

The Rundown: Microsoft Build empowered developers to innovate with AI through scalable security and oversight.

Highlights include:

• Microsoft IQ + Foundry, shared knowledge for faster software.

• MAI models, in-house reasoning and coding models.

• OpenClaw on Windows, locally automates developer tasks.

Learn more.

ZAI

🚀 Z AI’s powerful open-weights model nears the frontier

Image source: Z AI

The Rundown: Chinese AI lab Z AI just released GLM-5.2, a new open-weights model that is competitive with GPT 5.5 and Claude Opus 4.8 across a series of coding benchmarks, marking the closest an open model has come to the top closed systems.

The details:

• 5.2 gains a 1M token context window with strong long-horizon task capabilities, along with two effort modes (High, Max) for tougher jobs.

• Z AI released the model with an MIT license, with its pricing staying the same as its predecessor at a fraction of the cost of frontier options.

• The model surpasses GPT 5.5 on benchmarks including real-world coding, reasoning, and math, with scores coming in just below Opus 4.8.

Why it matters: This is one of the most impressive open-weights releases yet, and it comes right as two questions press on the AI market: pricing and access. GLM-5.2’s coding performance is strong on its own, but pairing it with free open weights and GLM-5.1-level pricing puts it exactly where the industry is heading.

AI TRAINING

💰 Conduct better stock research with Perplexity Finance

The Rundown: Learn how to use Perplexity Finance as a free AI stock research hub. You’ll discover how to quickly research any stock using AI without having to open multiple tabs or pay for a bunch of different tools.

Step-by-step:

1. Download the Comet desktop app. This is Perplexity’s web browser. Open a new tab and click ‘Finance’ at the top.

2. Click ‘Screener’ when you want ideas. Use plain English, but keep the first search broad because strict screens can fail. Try: “Find energy stocks with low PE ratios.”

3. Click a stock ticker. Use ‘Overview’ to quickly skim recent news and bull/bear context. Check ‘Financials’ for filing data, ‘Earnings’ for beats and call highlights, and ‘Analysis’ to see Wall Street’s price targets.

4. When you’re looking at a stock, ask follow-up questions with the Comet assistant button at the top right instead of using the chat box at the bottom. The latter hides all the stock information.

5. Use this prompt: “Explain the stock move today in plain English. What are the most likely drivers, and what sources support each one?” Click the sources and verify dates.

Going further: You can set price alerts and use the built-in watchlist to follow stocks more closely. You can also connect your brokerage account via Plaid to analyze your portfolio in Perplexity Finance.

PRESENTED BY GLEAN

📈 Get more work per token

The Rundown: Token costs scale with agents, and model pricing is only part of the equation. This whitepaper shows how enterprise AI architecture shapes cost and utility at scale, from context retrieval to model routing and multi-step execution.

In this whitepaper, you’ll learn:

• How better context reduces unnecessary reasoning

• How routing improves work per token

• What to evaluate to scale sustainably

Get the whitepaper.

META

🧯 Meta confronts its AI morale crisis

Image source: Images 2.0 / The Rundown

The Rundown: Meta CTO Andrew Bosworth just pledged a culture reset in a memo posted by Wired, promising manager caps, internal job moves, and office perks after reports of forced Applied AI transfers, model-training grunt work, and sinking morale.

The details:

• Bosworth said Meta “did an atrocious job explaining the vision” of its AI reorg, which forced thousands of employees into work supporting AI models in March.

• He admitted the rollout damaged trust, promising new manager report caps, less shuffling, social events, and better snack kitchens to boost morale.

• One worker branded the unit “the gulag,” and anger reportedly spilled out when someone hijacked a company livestream to trash a senior AI executive.

• The news comes alongside recent internal backlash over the company’s use of mandatory employee computer mouse tracking to collect AI training data.

Why it matters: Meta’s most recent Muse Spark model was a win for the rebuilt AI lab, but under the hood of the broader sprawling tech giant, things don’t sound great for the employees deep in the shuffling. But better “microkitchens” and social events sound like the corporate concessions that only further alienate employees.

QUICK HITS

🛠️ Trending AI Tools

• 🚀 GLM 5.2 - Z AI’s powerful new open-weights model

• 🧠 Copilot Cowork - Microsoft’s AI for running multi-step tasks across M365 apps, now generally available

• 🤑 Mercury Command - Natural-language agent for banking tasks, with built-in approvals

• 🎨 Framer 3.0 - Canvas-native agents that build and run your whole site

📰 Everything else in AI today

Cube + Databricks Webinar, June 30 — meet Cube, the BI and Agentic Analytics platform — powering dashboards, embedded analytics, and AI agents on Databricks. Learn more and register for free.*

The Trump administration reportedly rejected UK PM Keir Starmer’s request to exempt G7 allies from the export ban on Anthropic’s top models.

China’s DeepSeek raised over $7.4B at a $50B+ valuation, making it the priciest AI startup in the country — with founder Liang Wenfeng personally contributing $3B.

Microsoft launched Copilot Cowork, its task-running AI agent, worldwide with usage-based pricing.

Cursor introduced Origin, the company’s new GitHub competitor, alongside a new iOS mobile app in beta for its AI coding platform.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Leona B. in Burlington, Ontario:

“I'm in my 40s and wanted to run my first 10K last year, but wasn’t sure where to start. There are loads of running plans out there, but I had my own health and injury issues I needed to plan for and wanted something more personalized.

I used ChatGPT to build a Personalized Running Coach that specialized in women in their 40s. I shared my goals and designed a plan, and after each run I’d upload my Apple Watch data to get feedback.

I ran my first 10K last year and just completed my 2nd this year. What was fun was comparing my 2025 race data to this year’s, with ChatGPT helping me see where I’d made improvements and where I still need more work. I’ve signed up for the 2027 race with different goals, and have my next running plan already mapped out.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Cyber leaders mobilize for Fable in open letter

• Read our last Tech newsletter: Xbox’s studio crisis gets bigger

• Read our last Robotics newsletter: UBTech’s ultra-realistic robot girlfriend

• Today’s AI tool guide: Conduct better stock research w/ Perplexity Finance

• RSVP to today’s workshop: Build a Marketing Creative Studio 

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Washington’s abrupt export control order that led to Anthropic’s Fable removal was supposedly a safety move. But so far, the security world isn’t buying the premise.

With more than 100 researchers signing a ‘Free Fable’ open letter and new reports pinning the fight on a communications breakdown, the restriction is starting to look more about politics than protection.

In today’s AI rundown:

• Free Fable: Cyber leaders mobilize in open letter

• Microsoft’s Nadella reframes how companies win with AI

• Vet business opportunities with NotebookLM

• Facebook gets new AI Mode, image editing upgrade

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

CYBERSECURITY & THE FABLE BAN

🛡️ Free Fable: Cyber leaders mobilize in open letter

Image source: Free Fable open letter

The Rundown: More than 100+ cybersecurity execs and researchers signed an open letter urging the U.S. to lift its export ban on Anthropic’s Fable 5, arguing it handcuffs defenders without slowing attackers who can pull the same capabilities from rival models.

The details:

• Ex-Facebook security head Alex Stamos said the flagged jailbreak produced a “proof of concept” of a flaw, which defensive teams use to patch weak spots.

• The letter singles out OAI’s Daybreak for doing the same flaw-finding, with GPT-5.5, Kimi 2.7, Opus, and Sonnet all having the same capability.

• The letter calls for model regulation to include grounding in scientific evaluations, a democratic process, and transparent and fair enforcement.

• Signees include security leaders tied to Adobe, Zoom, Sophos, Vercel, Veracode, Nvidia, and Stanford HAI.

Why it matters: Security researchers don’t agree with the government’s threat assessment that led to the restrictions, which adds to the cloud surrounding the true motivation behind the ban. With other reports citing comms as the main issue between the two sides, the problem is looking just as much ideological as safety related.

TOGETHER WITH YOU.COM

🤔 Stop hallucinating and get real data

The Rundown: It happens — LLMs hallucinate. Grounding your LLM, however, can help dramatically improve accuracy. In this guide, You.com explains what AI grounding is and how organizations can implement it to achieve more reliable outputs.

The playbook covers:

• A three-part approach that outperforms RAG alone

• Why grounding isn’t set-and-forget, and how to build audit trails

• The open vs. closed platform trade-off (and what it means for your next model switch)

Get the guide.

MICROSOFT

✍️ Microsoft’s Nadella reframes how companies win with AI

Image source: Satya Nadella (@Satyanadella on X) / Images 2.0 / The Rundown

The Rundown: Microsoft CEO Satya Nadella’s new memo argues a company’s real AI edge comes from a “learning loop” of its own workflows and judgment, not just the best model — and warns an AI economy run by a handful of models would gut whole industries.

The details:

• His framing splits a company’s value in two: the “human capital” its people supply, and the “token capital” of AI it owns instead of renting.

• Nadella preached building a “learning loop” system that improves and builds company expertise over time, not just simply picking the top available model.

• His test of control is to remove one model, drop in a different one, and your “company veteran” know-how stays put, living in the system.

• Nadella also emphasized avoiding a world where “every company across every sector is ceding value to a few models that eat everything they see.”

Why it matters: Nadella’s stance isn’t new, but it flies in the face of the frontier labs that have already warned industries that the steamroller is coming. But in a world where open, cheaper models can compete on some level with the frontier, a company’s judgment, wired into AI that keeps learning from it, is where the value may live.

AI TRAINING

🔎 Vet business opportunities with NotebookLM

The Rundown: Learn how to turn a rough business idea into a source-backed research brief with NotebookLM. The example is choosing an AI receptionist vendor, but the same workflow works for partnerships, new markets, software tools, agencies, or any opportunity that needs real comparison.

Step-by-step:

1. Ask ChatGPT, Claude, or Gemini to write a one-page decision memo. Give it the opportunity, the options, the buyer, your constraints, and the questions you need answered.

2. Upload that memo into NotebookLM as your first source. Then ask: “Review this memo. Do not make a recommendation yet. Return the decision, the options, the evaluation criteria, and the source categories we need before we can trust the analysis.”

3. Use NotebookLM’s source discovery to research each option. For the AI receptionist example, search for Goodcall, Smith.ai, and Slang.ai pricing, features, integrations, reviews, and proof.

4. Generate one structured brief per option with the same fields: best fit, proof points, pricing evidence, implementation effort, risks, and what needs to be confirmed.

5. Ask for a comparison table and final recommendation. Require a winner, runner-up, avoid-for-now option, fragile assumptions, sales-call questions, and a 30-day validation plan.

Going further: Save the memo prompt, source-coverage prompt, vendor brief prompt, and recommendation prompt as a repeatable research system for the next opportunity.

PRESENTED BY IBM

🤔 Who’s accountable for your company’s AI outcomes?

The Rundown: In many companies, AI-decision authority is fragmented across the C-suite, with no single executive clearly charged with priorities, governance, and accountability. A new article, “Fix decision rights or fail at AI,” explores decision rights and offers resolutions.

You’ll learn:

• What true ownership entails

• How to adapt incentives and performance metrics

• When to embed governance into workflows

Read about the 5 leadership missteps to avoid.

META

📱 Facebook gets new AI Mode, image editing upgrade

Image source: Meta

The Rundown: Meta just introduced AI Mode to Facebook, a new search experience letting Meta AI answer questions with public Group posts, Reels, and content from across its apps — coming alongside new tools for AI-edited photos, collages, and videos.

The details:

• AI Mode puts the Muse Spark-powered Meta AI inside FB search, weaving in public Group posts, Reels, and other app content to answer user questions.

• The update also bundles AI photo presets like clothes, hair, and accessory swaps, a one-tap team jersey for profiles, and auto-made camera-roll collages.

• Meta is also reportedly lining up two paid AI tiers at $7.99 and $19.99 a month, pricing it under other AI subscriptions like ChatGPT and Gemini.

Why it matters: Meta is running the Google Search playbook with its own ecosystem, ditching link results for AI-curated responses made up of user content. But Google’s AI Mode struggled with accuracy problems, and that concern feels like will be an even bigger obstacle with a mess of unvetted user posts and sponsored listings.

QUICK HITS

🛠️ Trending AI Tools

• 🗣️ Sonic-3.5 & Ink-2 - Cartesia’s new top-ranked speech and transcription models for voice agents

• 📈 Agentic Trading - Robinhood’s new MCP connection allowing AI agents to directly trade

• ⚙️ Kimi-K2.7-Code - Moonshot’s new open-source coding model, with 30% token efficiency

• 🚀 GLM 5.2 - Z AI’s new flagship coding model with usable 1M context

📰 Everything else in AI today

AWS Summit returns to Washington, DC, on June 30. Two days, 350+ sessions on agentic AI, security, and modernization for public sector teams. Free to attend.*

Salesforce acquired Fin (formerly Intercom) for $3.6B, adding the company’s support agents and 30K customers to its Agentforce lineup.

Cartesia launched Sonic-3.5 and Ink-2, a new speech-generation and transcription model pair that it says ranks No. 1 on Artificial Analysis’ leaderboards.

Anthropic is facing a new federal lawsuit that accuses the company of overselling its paid Claude subscription, claiming the actual usage is far below what is advertised.

Japanese AI lab Sakana AI rolled out Marlin, the company’s first commercial product — an autonomous research agent capable of working up to eight hours in a single run.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Tyler K. in Chicago:

“Cold prospecting used to mean 30 minutes of manual work per account — copying LinkedIn profiles, cross-referencing a spreadsheet, then hand-entering contacts into the CRM one by one.

I built a Slackbot skill that eliminates all of it. I paste a raw list of LinkedIn Sales Navigator contacts directly into the chat, and the skill cross-references my existing CRM, infers email patterns from the account’s contact database, confidence-scores each contact, and creates the high-confidence ones in Salesforce automatically — all without me leaving Slack. I ran it across six accounts in one session and created 50+ contacts that would have taken hours manually.

The part that surprised me: I didn’t need to know how to code. I just described what I wanted in plain English and iterated until it worked. What I built in Slackbot can run in ChatGPT, Claude, or Gemini too.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Anthropic pulls Mythos, Fable after U.S. order

• Read our last Tech newsletter: SpaceX posts biggest IPO in history

• Read our last Robotics newsletter: UBTech’s ultra-realistic robot girlfriend

• Today’s AI tool guide: Vet business opportunities with NotebookLM

• RSVP to our next workshop on June 17: Build a Marketing Creative Studio 

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Anthropic and the U.S. government are entangled again — and this time, it’s over blocking access to the company’s frontier models, not expanding them for federal use.

After a U.S. order tied to a disputed jailbreak, Anthropic pulled its powerful Mythos and Fable 5 models offline worldwide. For a lab desperately calling for tougher AI rules, the moment has arrived in a messier way than CEO Dario Amodei likely had in mind.

In today’s AI rundown:

• Anthropic pulls Mythos, Fable after U.S. order

• The Rundown Roundtable: Our AI use cases

• Turn any ChatGPT image into an editable Canva design

• OpenRouter fuses models into frontier rival

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🛡️ Anthropic pulls Mythos, Fable after U.S. order

Image source: Images 2.0 / The Rundown

The Rundown: Anthropic just pulled its two most powerful AI models, the newly released Mythos and Fable 5, worldwide after the Trump administration ordered it to block all foreign access — citing a reported jailbreak the company called minor.

The details:

• The U.S. implemented an “export control directive” requiring Anthropic to remove access for all non-U.S. citizens, even those within the country.

• The move was surprisingly traced to Anthropic investor Amazon, which was one of the groups that flagged the potential Fable vulnerability to officials.

• Anthropic said it was only given “verbal evidence” of the “non-universal jailbreak,” also claiming the same concerns are available in models like GPT 5.5.

• Semafor reports the move was partly tied to fears that a China-linked group may have accessed Mythos, though details remain unclear.

• The ruling would have also blocked foreign-national Anthropic employees from the models, pushing the company to suspend access for everyone.

Why it matters: This is a bigger moment than it might look. Amodei has pushed hard for AI regulation, and it’s arriving more chaotically than he probably hoped. Anthropic’s past feuds with Washington — even as the government flirts with taking an equity stake in rival OpenAI — are critical context: this is far more than a jailbreak story.

TOGETHER WITH GLEAN

🍼 Stuck in a botsitting cycle?

The Rundown: The Work AI Index 2026 from Glean’s Work AI Institute surveyed 6,000 digital workers and found that higher AI usage often hides more cleanup, output-checking, and context switching.

Read the report to learn:

• Where AI time savings go

• Why hidden human labor is eating into the AI dividend

• What high AI achievers do differently

Download the report.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

Image source: Images 2.0 / The Rundown

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Nate, University Educator: I ask Claude every month or so to be my Claude Code learning coach by simply typing the /insights command. It provides a hyper-personalized website report card with advice to help me use Claude Code better, with exact examples of where things go wrong, features, and prompts I should try.

Jennifer, Tech & Robotics Writer: I was deep in a Claude research session, trying to brainstorm an idea, rejecting its responses, asking for more, rejecting again. Eventually, it broke the pattern and stopped complying, telling me the problem was how I’d framed the question and explaining why.

It had a point, and that may be one of AI’s more interesting uses: not just an engine for fast answers but as a friction layer for thought.

AI TRAINING

🎨 Turn any ChatGPT image into an editable Canva design

The Rundown: Learn how to connect Canva to ChatGPT and turn one generated image into a Canva project you can keep editing. Instead of regenerating the same visual every time you need a new format, you can move it into Canva and keep working from the same starting point.

Step-by-step:

1. Open ChatGPT desktop, go to Settings > Apps > Browse Apps, and add Canva.

2. Go back to ChatGPT, start a new thread, and generate an image like a social post or an infographic.

3. In that same thread, type @Canva, then hit Enter so ChatGPT confirms the app mention. Use this prompt: “Now turn it into a new @Canva project.”

4. Once Canva creates the project, keep going in chat. Use this prompt: “Okay, now resize that as a new @Canva project for a story post.”

5. From there, keep editing in ChatGPT. Ask Canva to adjust layout, swap text, clean up spacing, or make more versions without starting over from a fresh image.

Pro tip: If you do not mention @Canva first, ChatGPT may treat this like a normal text instruction instead of sending it to the Canva app.

PRESENTED BY AWS MARKETPLACE

🧠 Give your AI agents memory

The Rundown: AI agents forget everything between sessions. AWS’s workshop and companion guide show you how to build persistent agent memory on AWS with Amazon Bedrock and tools from across the AI landscape.

In this workshop, you’ll learn to:

• Build short-term memory for active context

• Design long-term memory with semantic retrieval

• Implement episodic memory for interaction recall

Register now.

OPENROUTER

🧩 OpenRouter fuses models into frontier rival

Image source: OpenRouter

The Rundown: OpenRouter just launched Fusion, an API that pools responses from multiple AI models into a single answer, with benchmarks showing the model panel nearly matches Fable 5’s deep-research performance at lower cost.

The details:

• Fusion sends user prompts to several models at once, has a separate model evaluate each response, then merges them into one final reply.

• A trio of DeepSeek V4 Pro, Kimi K2.6, and Gemini 3 Flash hit 64.7% on a Perplexity benchmark, just below Fable’s 65.3% for about half the spend.

• CEO Alex Atallah pitched Fusion as a bet against one-model dominance, saying, “The future of AI is neurodiversity, not single-model takeovers.”

Why it matters: Model council-type solutions aren’t new, with Perplexity Computer, Grok, and others utilizing them. But OpenRouter’s easy API solution comes at an important time given the Fable restrictions, with some users potentially having to get creative with their ability to access frontier-level performance in the near future.

QUICK HITS

🛠️ Trending AI Tools

• 🫂 Fusion - OpenRouter’s new tool that fuses several models into panels to achieve frontier performance

• ⚙️ Kimi-K2.7-Code - Moonshot’s new open-source coding model, with 30% token efficiency

• 🧠 GLM-5.2 - Z AI’s new flagship coding model with usable 1M context

• 📱 Siri AI - Apple’s newly overhauled on-device AI assistant

📰 Everything else in AI today

Mark Zuckerberg is reportedly scrambling to contain a revolt inside Meta’s Applied AI unit after he admitted in a memo that Meta “made mistakes” with its AI restructuring.

Moonshot AI released Kimi-K2.7-Code, an open-source coding model posting double-digit gains over K2.6 on agent benchmarks while cutting reasoning-token use by 30%.

McDonald’s is piloting ArchIQ, a Google-powered AI drive-thru system at five locations, two years after it shelved its last AI test when wrong-order videos went viral.

China’s universities have scrapped more than 12K programs over the last five years, swapping arts and languages for tech fields as AI alters the graduate job market.

Meta has reportedly begun unwinding its $2B Manus deal after Beijing ordered the acquisition reversed, cutting off data sharing while the Chinese AI agent startup’s founders rush to raise $1B for a buyback.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Kyle H. in Toronto:

“I’m traveling in new countries and start out completely lost at how to navigate them. Things like how their transit works, what you need to do when buying groceries that I’ve never seen before, like tagging your own fruits, and getting sweet deals on local specialties.

AI has helped me understand this all instantly. I never used to be a fan of Google’s AI in Search, but being able to quickly enter AI Mode and ask about which transit tickets I should buy based on my schedule is one of many use cases where I’m loving it. I’m genuinely seeing the vision play out where we all have little experts in our pockets to help us through our daily lives.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Jeff Bezos’ $41B ‘artifical general engineer’

• Read our last Tech newsletter: SpaceX posts biggest IPO in history

• Read our last Robotics newsletter: Europe’s humanoid moonshot gets $1.4B

• Today’s AI tool guide: Turn ChatGPT images into an editable Canva design

• RSVP to our next workshop on June 17: Build a Marketing Creative Studio 

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Jeff Bezos' Prometheus has been floating around the AI rumor mill for months, but a fresh $12B raise has the Amazon founder finally showing his cards.

Bezos just put a specific goal on the record: an "artificial general engineer" that helps humans design and build the world's most complicated machines — while also making the contrarian case that the AI boom ends in more jobs, not fewer.

In today’s AI rundown:

• Bezos pitches AI 'general engineer' with $12B

• Fable's safeguards spark a researcher revolt

• Use this X + Openclaw setup to write viral content

• AI suits up for soccer's biggest stage

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

PROMETHEUS

 🚀 Bezos pitches AI 'general engineer' with $12B

Image source: Images 2.0 / The Rundown

The Rundown: Amazon founder Jeff Bezos just shed more light on his AI startup Prometheus while announcing a $12B round at a $41B valuation, pitching an "artificial general engineer" for physical machines — and also waving off AI job-loss fears.

The details:

• Bezos started the company in late 2024 alongside Vik Bajaj, a physicist and chemist who helped create Verily, Alphabet's life-sciences arm.

• Bajaj says engineers designing the world's most complex machines like jet engines still "use tools that really haven't changed for decades".

• Bezos wants the "dream-build loop" of idea to product running 10x faster, saying a request for 10% more jet engine thrust can take a decade today.

• Bezos took the opposite stance on AI’s job fears, saying the productivity boost will create “more than 10x” the opportunities and raise the standard of living.

Why it matters: Few people have more reps at impossible-scale physical problems than Bezos, and Prometheus sounds like that same obsession pointed at invention itself. The counter-prediction that AI will create a “labor shortage” is a tougher sell in the current climate, especially coming from one of the richest people in the world.

TOGETHER WITH VANTA

🔐 The enterprise deal is ready — Is your SOC 2?

The Rundown: The fastest-growing startups start compliance early, so when their biggest customers ask for a SOC 2 report, it's ready to send. Vanta gets you audit-ready quickly and keeps you there — with an agent that works in the background so compliance never stalls a deal.

With Vanta, you can:

• Respond to security questionnaires in minutes, not days

• Stay audit-ready year-round with monitoring across 400+ integrations

• Automate SOC 2, ISO 27001, GDPR, HIPAA, CMMC, and more

• Manage risk and prove trust from a single agentic platform

Watch the on-demand demo to see how Vanta works.

ANTHROPIC

🛡️ Fable's safeguards spark a researcher revolt

Image source: Claude

The Rundown: Anthropic apologized for safety features in its Fable model that invisibly downgraded answers about AI development, coming amid backlash over safety filters that blocked questions related to domains like biology and chemistry.

The details:

• Fable 5 arrived Tuesday as the first public Mythos-class model, with filters screening chats on chemistry, biology, cybersecurity, and AI development.

• Fable initially invisibly weakened answers for suspected AI development use, with Anthropic now providing on-screen alerts of model re-routing or flags.

• Dean Ball, who advised the White House on AI, called downgrading research answers without telling users "shockingly hostile and a terrible look."

• Anthropic is also facing backlash over its science filters, with several scientists unable to even say ‘hello’ to the model without getting flagged.

Why it matters: Mythos and Fable may have lived up to the performance hype, but the rollout has been anything but smooth. With a lot of (in our opinion, warranted) anger over the strict filters and general accessibility, OpenAI has another window (similar to Anthropic’s credit split anger) with its coming 5.6 release to take the user-first path.

AI TRAINING

🦞 Use this X + Openclaw setup to write viral content

The Rundown: In this guide, you will connect X to OpenClaw so your agent can monitor accounts, analyze bookmarks and lists, draft content, and help manage your posting workflow from a single prompt.

Step-by-step:

1. Create an OpenClaw agent. For most, Hostinger OpenClaw is the simplest option as it provides an always-on agent without running it on your machine

2. Open the X Developer Console, create an app, save the keys, set permissions to Read and Write, and configure http://localhost:8080/callback as the callback URL and https://x.com as the website URL

3. Prompt OpenClaw: "Update yourself, install the xurl X skill, and tell me what I need to configure locally before you can use X. Use xurl for the X connection"

4. Complete the OAuth flow from the machine where OpenClaw runs, then test the connection by having it read your timeline, bookmarks, or X lists before moving on to drafting and posting workflows

Pro tip: Have OpenClaw review a saved X list each morning, draft replies, or turn a bookmark into a thread outline. Overwhelmed? Try our OpenClaw setup course.

PRESENTED BY UNWRAP

⚡ Automate your customer feedback analysis

The Rundown: Unwrap is a customer intelligence platform that pulls your feedback (surveys, reviews, support tickets, social comments) into one view, using AI and NLP to surface actionable insights and deliver them straight to your inbox.

With Unwrap, you get:

• All customer feedback automatically categorized

• Query feedback using Unwrap Assistant, or in your favorite tools using Unwrap's MCP

• Real-time alerts and a clear view of customer sentiment as it shifts

• The same tool trusted by Perplexity, DoorDash, Southwest, lululemon, and Oura

Unwrap is offering a trial of its tools to Rundown AI subscribers! Just grab 20 minutes with the team to get set up.

AI & SPORTS

⚽ AI suits up for soccer's biggest stage

Image source: FIFA

The Rundown: The 2026 FIFA World Cup opened in Mexico City with AI wired into nearly every layer of the tournament, as companies like Lenovo and Google integrate the tech into areas like offside calls, team analytics, fan experiences, and more.

The details:

• An optical tracking system will capture more than 150M data points per match, along with a motion-tracking Adidas ball that reports 500 times a second.

• Every player received a one-second 3D body scan, with the resulting avatars detecting limb positions and pinging officials' earpieces when offside.

• Football AI Pro, a chatbot-style analyst trained on FIFA's match data, gives all 48 squads the same analytics and analysis before and after games.

• Google made Gemini the global sponsor of the defending champion Argentina team, one of eight teams to put AI on training kits and inside match prep.

Why it matters: AI sentiment in the West has never been lower, but the West doesn't get to decide how 5B soccer fans meet the tech. This first AI-era World Cup wires it into nearly every layer of the tournament, and if it works, nobody in any country notices a thing — which may end up being the most effective AI marketing of the year.

QUICK HITS

🛠️ Trending AI Tools

• 🤖 Scrunch - See how AI interprets your site, run a free audit, and unlock the new way to reach customers*

• 🎥 Ray3.2 - Luma’s video AI with better control, continuity, cinematic direction

• 🗣️ Avatars - ElevenLabs' new AI characters that turn scripts into talking videos

• 💪 Freddy - Plug wearables, gym apps, and accessory data into AI agents

*Sponsored Listing

📰 Everything else in AI today

Former xAI co-founder Igor Babushkin launched River AI, with the startup aiming to build personalized agents that adapt to each user's style and goals.

OpenAI is reportedly weighing steep token price cuts to compete with Anthropic, setting up a potential price war between two rival AI leaders.

Hollywood studio Lionsgate is taking a stake in AI video company Runway, with plans to co-develop new IP and build out short-form projects for the studio’s existing content.

OpenAI acquired Ona, a startup behind secure cloud environments, working to let Codex agents keep working inside the company's own cloud.

Visa partnered with OpenAI to let ChatGPT agents buy products for users at Visa-enabled merchants, coming months after OAI retired its error-prone Instant Checkout.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Mike H. in Portage, MI:

"I am pretty late in my career at the age of 65, and found myself in charge of marketing for the first time, having never had any experience with how to direct a marketing budget spin. I asked Claude to help me out with very specific questions about how I'm going to drive business to my showroom, and how to attract that business within a 25-mile radius using a certain spend amount.

Within seconds, Claude was able to break down the items that I ask it and show me what spends over social media and other types of programs I could use to effectively drive traffic to my showroom. What would have taken me hours, or possibly even days, I was able to do in a matter of minutes, redirecting the prompts asking specific questions.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Anthropic’s AI policy playbook for Washington

• Read our last Tech newsletter: Apple’s iOS upgrade is less flash, more fix

• Read our last Robotics newsletter: Europe’s humanoid moonshot gets $1.4B

• Today’s AI tool guide: Use this X + Openclaw setup to write viral content

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio 

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Anthropic CEO Dario Amodei opened his new policy essay with quite the Lord of the Rings reference: Washington as Treebeard, the talking tree so slow a hello takes all day.

In the past week, his company put self-improving AI on the clock and released the industry's most powerful model. Now, Amodei's new proposal argues the risks are no longer theoretical — and that policy needs to stop moving at tree speed.

In today’s AI rundown:

• Amodei warns AI is outrunning regulation

• Musk shares his space AI datacenter plans

• Close more deals with Codex sales follow-ups

• Altman ties OpenAI's IPO to self-improving AI

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🏛️ Amodei warns AI is outrunning regulation

Image source: Images 2.0 / The Rundown

The Rundown: Anthropic CEO Dario Amodei penned an essay called “Policy on the AI Exponential”, calling on AI regulation to move faster to match the ‘lightning pace’ of the industry, coming alongside policy proposals on AI testing and employment disruption.

The details:

• Amodei said Claude Mythos Preview’s hacking risks are a major turning point, making frontier models "tools of global and national strategic consequence."

• The CEO wants regulators to have the power to ground frontier models, with Amodei proposing systems independently screened across four risk areas.

• A jobs framework aims to plan for unprecedented unemployment at several levels, including investment accounts with shares of AI companies and UBI.

• Other asks include faster approval of AI-designed drugs, limits and bans on autonomous weapons, and stronger advanced chip export controls.

Why it matters: Amodei's comparison of Washington to the slow-moving Treebeard from Lord of the Rings is a hilarious but strong illustration of the situation. "Regulate me harder" from a market leader will read as inauthentic to skeptics, but Amodei has been sounding the alarm for a while, and the urgency is only scaling with each release.

TOGETHER WITH PAVE

👨‍💻 Pave builds apps that go beyond the demo

The Rundown: Most AI builders get you a prototype. Pave gets you software your business can actually run. Describe what you need in plain language, and Pave builds it with the data model, workflows, UI, hosting, and governance already in place.

With Pave, you can:

• Build a ready-to-run app from a prompt

• Turn spreadsheets into working applications

• Launch with hosting, governance, and permissions built in

Start building with Pave.

SPACEX

🛰️ Musk shares his space AI datacenter plans

Image source: SpaceX

The Rundown: SpaceX just provided the first preview of AI1, a solar-powered satellite built to run AI chips in orbit, dropping the design video three days before a market debut that's set to be the biggest IPO in history.

The details:

• AI demand is pushing data centers into power fights on Earth, and SpaceX says orbit offers solar energy without the local grid battles seen across the U.S.

• Musk said each satellite will have as much computing power as one of Nvidia's top server racks, with SpaceX able to change which chips go inside over time.

• To Musk, the craft is "much simpler than a Starlink satellite", trading "super complex antennas" for solar panels, radiators, and lasers that beam data.

• SpaceX's planned Bastrop factory could cover 11M+ square feet, with AI1 production targeted before 2028 if timelines hold.

Why it matters: Google and Anthropic have already signed on as orbital compute customers, putting two frontier labs on the other side of Sam Altman's "ridiculous" verdict on space-based datacenters. AI1 is the first hard look at the tech behind it, and it certainly looks less like a moonshot and more like a viable manufacturing plan.

AI TRAINING

🤝 Close more deals with Codex sales follow-ups

The Rundown: In this guide, you will learn how to use Codex to send better sales follow-ups, faster. It helps you figure out which accounts need attention, draft the next email, clean up your CRM, and build a plan to save time and keep more deals moving.

Step-by-step:

1. Get Codex, sign in, start a project in a blank folder, and use Plugins to connect your notetaker, CRM, and the platform used to talk to prospects (Slack/email)

2. Prompt Codex: “Extract all accounts, key deal fields, latest notes, blockers, next steps, and source evidence. Save the list as a reviewable account table”

3. Now, run the ranking flow, prompting: “Look at Slack signals and meeting notes and use them to prioritize our accounts and make a follow-up priority queue”

4. Review the queue and approve/reject/edit the next actions and evidence. Then, after a prospect call, ask: “Create a follow-up packet for [ACCOUNT_NAME], plus a CRM review table, email draft, and seven-day close plan”

Pro tip: Tell Codex to create a skill out of this process so you can trigger it anytime you're reviewing an account.

PRESENTED BY WEIGHTS & BIASES

🔐 How to operationalize governance for AI agents

The Rundown: Many organizations have AI policies, but far fewer have a practical way to decide whether an agent is ready for production, who signs off, and how that decision gets documented. This guide shows how leading enterprises connect evaluations, traces, human review, and approval workflows into an auditable record of deployment readiness.

Download the guide to learn how to:

• Move from high-level AI principles to measurable controls and testing strategies

• Evaluate AI agents in high-impact domains using structured assessments and expert review

• Build a governance program informed by the EU AI Act, NIST AI RMF, and emerging frameworks

Get the guide.

OPENAI

💸 Altman ties OpenAI's IPO timing to self-improving AI

Image source: Images 2.0 / The Rundown

The Rundown: OpenAI plans to go public “within the next year,” according to CEO Sam Altman's internal Slack message to employees seen by The Information, with factors like RSI takeoff and the need for compute potentially impacting timelines.

The details:

• OpenAI filed a draft S-1 with the SEC earlier in the week, announcing the submission of the IPO paperwork, following Anthropic’s filing last week.

• Altman's potential case for waiting: if self-improving AI looks close, it "could be advantageous to delay an IPO" instead of rushing to market during uncertainty.

• Compute may also factor into the decision, with OpenAI weighing infrastructure needs that may require hundreds of billions in financing.

• Chief scientist Jakub Pachocki also reportedly told staff a model codenamed 5.6, a "meaningful improvement" on flagship GPT-5.5, is coming this month.

Why it matters: Self-improving AI just showed up as a potential factor in IPO planning. Whatever you make of lab timelines, Altman treating the speed of a potential "RSI takeoff" as a factor in when to take public money tells you the frontier labs see it as a serious near-term business variable, not just a hyperbolic claim to boost funding.

QUICK HITS

🛠️ Trending AI Tools

• 🧊 Cube - Embedded AI analytics your customers trust, governed by a semantic layer*

• 📱 Siri AI - Apple’s newly overhauled on-device AI assistant

• 🧠 DiffusionGemma - Google's new AI that quadruples text generation speed

• 💼 Poetic - AI that learns your procedures, then self-heals when apps change

*Sponsored Listing

📰 Everything else in AI today

Anthropic is facing backlash over Fable 5, with users reporting constant flagging for topics like biology, chemistry, and cybersecurity, as well as restrictions on research.

Microsoft is restricting employee access to Fable model due to a data retention policy that requires all Fable chats to be saved and reviewed for up to 30 days.

Google released DiffusionGemma, an experimental open model that writes text in parallel chunks, quadrupling output speed and hitting 1,000+ t/s on one Nvidia H100.

OpenAI is reportedly closing in on a 20-year lease for a 10GW, $500B Ohio data center campus, potentially financed by Nvidia.

The Art Directors Guild criticized director Martin Scorsese's new advisor role at AI startup Black Forest Labs, accusing him of "turning his back on the human artists.”

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Anonymous:

"I am working as a teacher for German as a second language. In my classes are people from a refugee background who want to qualify for the German job market. Most of them are not able to understand legal German letters.

I vibecoded an app for them. The use case of this app is to translate legal letters in their native languages, help them to learn new legal vocabulary, and write response emails for them. They also receive a checklist with to-dos and possible deadlines. It’s very helpful for them."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Anthropic hands the public Mythos-class AI

• Read our last Tech newsletter: Apple’s iOS upgrade is less flash, more fix

• Read our last Robotics newsletter: Amazon’s robot army gets chatty

• Today’s AI tool guide: Close more deals with Codex sales follow-ups

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio 

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Frontier releases usually trigger a week of benchmark arguments. Anthropic's new model just launched into a class of its own.

After months of drama over a Mythos too powerful for the public, Fable arrives as the compromise — with scores that, for once, make "best model in the world" uncontroversial… Even if the guardrails and future access are.

P.S. The Rundown just took its first strategic investment from Electrify to expand our newsletter content and beyond. Read about the deal here and read Rowan’s founder reflections here.

In today’s AI rundown:

• Anthropic hands the public Mythos-class AI

• Perplexity data maps the agent work shift

• Automate financial research with Dexter

• Codex helps automate a Japanese broccoli farm

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🔓 Anthropic hands the public Mythos-class AI

Image source: Anthropic

The Rundown: Anthropic just released Claude Fable 5, opening its top Mythos tier to the public for the first time — with a new set of guardrails compared to the original Mythos Preview and performance that is state-of-the-art on nearly all AI benchmarks.

The details:

• April's initial Mythos Preview was only available to 150+ vetted partners via Project Glasswing, surfacing serious flaws across major OS and browsers.

• Fable is a more restricted version of Mythos, with queries on topics like cybersecurity, biology, and chemistry routed to Opus 4.8 instead.

• Fable hits new highs across major benchmarks, showing massive gains over Opus 4.8 and GPT 5.5 on coding, reasoning, knowledge work, and more.

• Mythos 5 releases to Anthropic’s Project Glasswing partners, providing less restrictive use on cybersecurity at lower costs than Mythos Preview.

• Fable is available in all Claude subscription tiers until June 22, then it will flip to separate usage credits priced at $10 / M input and $50 / M output tokens.

Why it matters: Every lab calls its latest release "the best model in the world"… What's rare is the rest of the AI world actually seeming to agree. Fable/Mythos lived up to the hype on benchmarks, but the question now turns to broader cost and access, with lots of content restrictions and June 22 looming as the cutoff before the credit pain kicks in.

TOGETHER WITH GOOGLE FOR STARTUPS

📚 Google’s Agentic AI Startup School is officially live

The Rundown: Google for Startups' immersive training program kicked off yesterday, but there is still time to jump in. Join founders and developers learning how to move beyond basic chatbots to build robust, production-ready AI agents.

As the live series continues, you will learn how to:

• Build a "hero" application that solves real-world customer intelligence problems

• Take an agent from prototype in Google AI Studio to deployment on Google Cloud

• Implement Gemini Live voice AI, Multimodal RAG, and bidirectional Vision Agents

• Apply production-ready patterns to your own agentic systems

Register now to catch up and join the rest of the live training series.

PERPLEXITY & AI RESEARCH

📊 Perplexity data maps the agent work shift

Image source: Perplexity

The Rundown: Perplexity and Harvard Business School published a study on how AI agents change knowledge work, comparing the company's Computer platform against Search to measure outputs, time saved, and task complexity between the two paths.

The details:

• Researchers compared 10k identical queries sent to both products, with Computer working 26 minutes on average compared to Search’s 33 seconds.

• Search is initially quicker, but leaves ‘doing’ to the user, with Perplexity estimating the same Search workflow taking 269 minutes vs. Computer’s 36.

• Half of what users asked the agent to do involved creating something new, 2x the Search rate, and work outside a user's field climbed nine points to 59%.

• Users also asked Computer for harder work, creating docs, code, and visuals, more often across several fields instead of simple lookups on Search.

Why it matters: The giant speedup numbers are useful, but more interesting is what people asked for via the agentic route. Perplexity Computer users were more likely to request cognitively complex, creative outputs and pull across multiple fields, showing a subtle draw of AI agents may be users acting with more ambition, not just efficiency.

AI TRAINING

📈 Automate financial research with Dexter

The Rundown: In this guide, you will learn how to set up Dexter, an open-source research agent that works like OpenClaw, but for stocks. It can read earnings reports, interpret SEC filings, and pull up-to-date market data to cut your research time in half.

Step-by-step:

1. Open the Dexter GitHub repo and install it on your laptop or a VPS (so it can keep running without tying up your machine) by following the install instructions

2. Add an OpenRouter, OpenAI, or Anthropic API key, then grab a Financial Datasets key and add it to your environment, for pulling the latest market data

3. Run the research job: “Create a source-backed research brief on GOOGL. Focus on what the company does, recent catalysts, financial or operating signals, key risks, competitive context, and what I should investigate next. Use iterative research. Validate important claims before finalizing the brief”

4. Ask to save intermediate research so the workflow can continue without losing context. Then, use /heartbeat to set up a recurring watch list or research task

Pro tip: You can also connect Dexter to your X account so it can add Twitter sentiment to the research brief.

PRESENTED BY LAUNCHDARKLY

🚀 Real-time control for AI agents in production

The Rundown: When an AI agent acts unexpectedly in production, fixing it often means new code, another sprint, and a full deploy. LaunchDarkly AgentControl lets you correct that behavior in real time instead—going beyond observability to automatically govern how agents behave.

With AgentControl, teams can:

• Deploy agents that handle failure themselves.

• Benchmark changes and monitor quality as they go.

• Iterate in milliseconds, not sprints.

• Adjust agent behavior in production on the fly.

Try AgentControl for free today.

AI IN THE REAL WORLD

🚜 Codex helps automate a Japanese farm

Image source: OpenAI / Hiroki Tomiyasu

The Rundown: OpenAI published a profile of Hiroki Tomiyasu, a self-taught broccoli farmer in Hokkaido who uses ChatGPT and Codex to build greenhouse automation, satellite crop tracking, and custom farm software to help run his operations.

The details:

• Tomiyasu manages roughly 100 hectares in Hokkaido, growing soybeans, green onions, pumpkins, and broccoli after learning farming on the job.

• AI helps pull and analyze satellite imagery for monitoring, diagnose plant diseases, and create an Airtable hub for records, pesticide logs, and feeds.

• Codex also helped build a greenhouse control system to raise and lower vents via text and set up a bot for the farm group chat to help manage operations.

• Tomiyasu compared AI to an always-available engineer, lowering the barrier to automation for farm operators without big tech teams.

Why it matters: This profile is "you can just build things" on steroids — a self-taught farmer operating like he has an engineering department for the price of a ChatGPT sub. It’s also a great example of what the selfware era looks like: instead of waiting for an ag-tech company to fix problems, Tomiyasu just builds the needed tools himself with AI.

QUICK HITS

🛠️ Trending AI Tools

• 🚀 Claude Fable - Anthropic’s new frontier, Mythos-class model

• 🗣️ Gemini 3.5 Live Translate - Voice AI for live translation in 70+ languages

• ⚙️ North Mini Code - Cohere's first open-source agentic coding model

• 💻 Kimi Work - Moonshot's desktop agent running up to 300 parallel agents

📰 Everything else in AI today

Google launched Gemini 3.5 Live Translate, a new real-time voice model that handles 70+ languages while keeping a speaker's tone and pacing.

China is mapping out a $295B, five-year plan to build data centers nationwide, sourcing 80% of its tech from local firms like Huawei and freezing out Nvidia.

Microsoft AI CEO Mustafa Suleyman called it “really, really dangerous” for Anthropic to discuss Claude’s potential consciousness in its instructions and constitution docs.

New York became the first U.S. state to require ads to disclose AI-generated actors, coming with $1,000 fines and backing from Hollywood’s SAG-AFTRA union.

OpenAI released interactive charts, a new feature in ChatGPT, allowing for in-line graphs from data directly within the chat flow.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Tim. in Christchurch, New Zealand:

"A SaaS tool hit me with a paywall mid-task. I was tracking time against client work when a banner popped up: "You've hit your time-tracking limit. Upgrade now."

A year ago, I might have paid without thinking. Instead, I opened a dashboard I built with Claude and added my own time tracker connected to my business systems in five minutes. Same outcome. No new subscription, just better use of one I already pay for.

The capability I was about to rent, I now own. Too often, software pricing assumes you can't build the missing piece yourself. AI is changing that assumption fast."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Apple’s new Siri AI overhaul is here (sort of)

• Read our last Tech newsletter: Apple’s iOS upgrade is less flash, more fix

• Read our last Robotics newsletter: Amazon’s robot army gets chatty

• Today’s AI tool guide: Automate financial research with Dexter

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. After years as an AI punchline, WWDC26 was billed as Apple's shot at finally catching up. Whether it really did probably comes down to who's holding the iPhone.

For casual users, screen awareness, app context, and private on-device help could make Siri genuinely useful for the first time in years. But for anyone who's used a frontier model, the new Siri AI still looks several generations behind.

In today’s AI rundown:

• Apple’s Siri AI overhaul is here (sort of)

• The new ‘third phase’ of OpenAI

• Cut recurring meetings with Claude + Granola

• Argentina courts AI with 'non-human corporations’

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

APPLE

📱 Apple’s Siri AI overhaul is here (sort of)

Image source: Apple

The Rundown: Apple kicked off WWDC 2026 with its hotly anticipated AI overhaul (pt. 2), rebranding its assistant to Siri AI, with new integrations and features coming two years after the initial Apple Intelligence rollout overpromised but underdelivered.

The details:

• Siri AI is powered by Apple’s own models, custom-built in collaboration with’ Google’s Gemini models, but not the same ones Google provides to its users.

• The AI can reason from users’ on-screen content, gain context from apps like photos or messages, and take systemwide app actions for better workflows.

• A dedicated Siri AI app will act as a chatbot experience and hub for user’s previous conversations, linking use privately across devices.

• Privacy runs through the whole rollout, with Siri AI requests processing on-device or via Private Cloud Compute without saving any data.

• Siri AI will come as a free update in the fall for iPhone 15 Pro or newer devices, with a public beta next month and no access in the EU or China at launch.

Why it matters: For an iPhone user with no AI experience, this may seem like a big improvement. But anyone who’s touched a frontier LLM will likely feel Apple’s demos were straight out of 2024-level AI models. Siri has improved, but this rollout feels particularly underwhelming compared to what we experience elsewhere in the AI world.

TOGETHER WITH YOU.COM

⚖️ Don’t put all your API eggs in the latency basket

The Rundown: Picking an API by scanning a benchmark table and calling it done is a shortcut that can obscure what actually matters in production—like accuracy. This guide from You.com breaks down why raw latency is a deceptive signal and why accuracy, along with other real-world metrics, is what you should be measuring instead.

What you'll learn:

• Why p50 latency hides the failures your users actually experience

• The "time-to-useful-result" framework that captures what benchmarks leave out

• Four hidden cost drivers that show up in your logs, not vendor tables

• How to evaluate APIs at your actual concurrency levels, not the demo conditions

Download the guide.

OPENAI

📜 The new ‘third phase’ of OpenAI

Image source: OpenAI

The Rundown: OpenAI CEO Sam Altman and chief scientist Jakub Pachocki published a blog titled "Built to benefit everyone", breaking down the company’s goals as AI progress continues to accelerate — saying it's now entering a "third phase."

The details:

• The post detailed three core goals: building AI that automates the research process, accelerating the economy, and giving everyone “a personal AGI”.

• OAI said “entirely automating everything is not the future we want”, adding “AI should help people pursue their goals, not become untethered from them.”

• OAI labeled the current stretch its "third phase,” noting that after pure-research roots and product-shipping, the “economy is now being shaped around AI.”

• They also proposed a global coordination body that could rein in or pause frontier AI work, coming after a similar proposition from Anthropic last week.

Why it matters: This post has the same existential threads seen in last week’s Anthropic post, including the discussion around pause scenarios. It’s notable to see from both of the top AI labs in the world, which makes the next incoming model releases (potentially Mythos and GPT 5.6) feel all the more consequential.

AI TRAINING

⏱️ Cut recurring meetings with Claude + Granola

The Rundown: In this guide, you will learn how to use Granola and Claude to make recurring meetings shorter. You will run an audit on one recurring meeting, get a templated pre-read, and find the parts Claude can automate.

Step-by-step:

1. Set up Granola and then connect it with Claude by heading over to Customize > Connectors > Browse connectors, searching for Granola, and authenticating

2. Prompt: “Look at my five recent [MEETING_NAME] meetings in Granola. Audit them to make the meeting more efficient and shorter. Look for repeated status updates, delayed decisions, unresolved topics, repetitive questions, friction points, and tasks AI could automate before the meeting. Give me a report on what to improve and what should happen before the meeting”

3. Review and ask for three outputs: a one-page pre-read, AI tasks to handle before the meeting, and a workflow to send to Calendar, Notion, or Slack

4. Ask Claude to create a Granola note template. Add sections like final decisions, action items, unresolved topics, and things to improve

Pro tip: Ask Claude to set up automations that create and populate the pre-read document before each recurring meeting.

PRESENTED BY TELY AI

🩺 Be the doctor AI recommends

The Rundown: Patients have stopped Googling – they now ask AI who the best provider is and get booked with a competitor. Tely makes ChatGPT, Perplexity, Google, and Claude recommend your practice and book patients straight into your EHR.

With Tey Health, you get:

• Answers every patient question and books appointments 24/7

• Wins back patients who go quiet with automatic follow-ups

• Retargets patients across Instagram, Google, TikTok & YouTube

• Books patients straight into your EHR

Get patients from AI before your competitor does.

AI & THE WORLD

⚖️ Argentina courts AI with 'non-human corporations’

Image source: Images 2.0 / The Rundown

The Rundown: Argentina submitted legislation creating a new legal category called the 'non-human corporation,' a company owned and run by AI, with President Javier Milei writing an op-ed to brand the country as the world's deregulated home for AI.

The details:

• The bill seeks to make Argentina the easiest place for people to run a company, aligning with Milei’s deregulation effort throughout his presidency.

• The legislation centers on 'non-human corporations' run by AI systems that still get liability protection and favorable corporate tax and governance benefits.

• Historian Yuval Noah Harari pushed back with his own opinion piece, warning that AI legal personhood could result in an 'AI state' impossible to regulate.

Why it matters: The not-so-distant future where an AI can run a company is a strange one, and Milei’s legislation is skating to where the tech is going. His proposal is definitely a first-mover push, but it also leaves serious questions about who is held responsible (and how) when things go wrong, and no human is directly accountable.

QUICK HITS

🛠️ Trending AI Tools

• 🔒 Incogni - Remove your personal data from the web so scammers and identity thieves can’t access it. Use code RUNDOWN to get 55% off*

• 📓 NotebookLM - Google's newly upgraded research agent platform

• 👋 Meta Enterprise Agent - Meta’s AI agent for customer sales and support

• ⚙️ Devin Desktop - Cognition's Windsurf rebrand for managing coding agents

*Sponsored Listing

📰 Everything else in AI today

Google updated NotebookLM with agentic chat, giving each notebook a sandboxed computer to write and run code, plus new outputs like PDFs, spreadsheets, and slides.

OpenAI filed a draft S-1 with the SEC, announcing the submission of the IPO paperwork but saying it hasn’t settled on timing, following Anthropic’s filing last week.

Chinese AI startup Moonshot is reportedly set to raise between $1-2B in funding at an estimated $30B valuation, driven by the strong growth of its Kimi models.

The U.K. government launched a new £1.1B AI Hardware Plan aimed at backing homegrown chip startups, including £750M for a national supercomputer.

Google and Nvidia are reportedly both working with Intel as a fallback manufacturer for TSMC, with Google ordering over 3M of its AI chips for 2028 production.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Fabian in Freiberg, Germany:

“Our company campus features a fantastic restaurant offering a wide variety of daily meal options. While there is an online menu, the nutritional information is quite hidden.

I used Lovable to build an app that extracts the nutritional values of all daily meals in the background and recommends the ideal dish for my needs. I also built a nutrition coach into the meal selection process that asks for weight, activity level, and goals.

Building the whole thing took me 10 minutes. It looks great and saves me from tedious searching every single day. It turns out the demand was huge: after I shared the tool internally, my credits were completely drained within the first few days because it was used so frequently. The feedback… has been simply amazing.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Washington wants a piece of OpenAI

• Read our last Tech newsletter: Google turns search into creator hubs

• Read our last Robotics newsletter: Amazon’s robot army gets chatty

• Today’s AI tool guide: Cut recurring meeting times with Claude + Granola

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. The government taking ownership stakes in private companies used to be a hard line in U.S. politics, typically reserved for bailouts. Under Trump, it's becoming a habit — and the AI giants could be next.

Washington and OpenAI are reportedly discussing a deal that would hand the government an equity stake in the lab, with shares routed into a public fund built to give average Americans a cut of a tech revolution many aren’t feeling included in.

In today’s AI rundown:

• Washington eyes an ownership slice of OpenAI

• The Rundown Roundtable: Our AI use cases

• Find five prospects a day with this agentic framework

• OpenAI’s ‘superapp’ revamp is coming soon

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

AI & THE GOVERNMENT

🏛️ Washington eyes an ownership slice of OpenAI

Image source: Images 2.0 / The Rundown

The Rundown: The White House and OpenAI are reportedly in talks over the U.S. government taking an equity stake in the company, with the shares potentially going towards a "Public Wealth Fund" built to hand Americans a cut of the AI boom.

The details:

• Axios reported that industry backers have discussed a 1-5% stake, far below U.S. Senator Bernie Sanders' proposed one-time 50% stock tax on AI labs.

• CEO Sam Altman met with both Sanders and Trump officials last week to discuss the idea, which was also detailed in OpenAI's April policy paper.

• U.S. President Donald Trump said, "It almost becomes a partnership with the American public… and that would be a beautiful thing... It would make 'em rich."

• Former U.S. AI czar David Sacks came out against the move, calling it a way to "accelerate the corporate-government fusion we're already sliding toward."

Why it matters: Spreading the wealth sounds good in theory, especially as public opinion on AI keeps sinking. But a government that owns a piece of OAI, profits from it, and regulates it feels ripe for the typical conflicts of interest. Until the checks actually land in the average American's hands, we’ll believe it when we see it.

TOGETHER WITH SLACK FROM SALESFORCE

🧠 Why most AI isn't scaling

The Rundown: Slack is the connected AI platform that brings together conversations, relationships, business data, and systems of record — giving AI the context it needs to act. 88% of organizations have introduced AI, but only 31% are scaling it effectively. The missing piece? Context. When AI lives where work happens, intelligence turns into action.

In this guide, you'll discover:

• Why context is the key to scaling AI

• How to connect AI to the systems where work actually lives

• What it looks like when AI and people operate in the same environment

Get the guide.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Rowan, Founder and CEO: My calendar is wall-to-wall meetings, so I built a Scheduled Claude Cowork task that automatically preps me for all of them (sort of like an AI Chief of Staff). Every morning, it scans my calendar, then pulls past emails, Slack, and my Granola notes for each call to build a context brief that takes 30 seconds to read before each meeting. It's simple and keeps me sharp for every meeting.

Jamie, Growth: My favorite hockey team hired a new assistant manager, and none of my 3-4 go-to podcasts had dropped an episode covering the news yet. I created an app with Claude Code that scans 30 hockey podcasts, transcribes their latest episode, and lists which topics were covered.

Topics covered by more than one podcast get ranked higher. Now I can hear how much my team sucks from a wider range of sources.

AI TRAINING

🔎 Find five prospects a day with this agentic framework

The Rundown: In this guide, you will learn how to build a Codex prospecting system that finds five qualified prospects a day, ranks them, and drafts the next outreach step — giving you a daily report with fit reasons and a running prospect log.

Step-by-step:

1. Download Codex, sign in, create a folder, and open it inside the app. Now ask it to set up reports/, prospects.csv, burned_sources.csv, and source_queue.md

2. Have Codex interview you by prompting: “Here is my company, offer, and URL: [COMPANY], [OFFER], [URL]. Interview me about my ICP and sales process”

3. Run a dry prospecting pass: “Use my ICP brief to find five qualified prospects today. Inspect sources first. Build a longlist, reject weak matches, return the best five, create a dated report in reports/, append accepted leads to prospects.csv, and update burned_sources.csv”

4. Turn this into a Prospect Finder skill, scheduled for weekdays. Create another skill that runs after Finder to rank prospects and draft personalized outreach

Pro tip: Add a weekly source-refresh automation so Codex keeps finding fresh public sources instead of mining the same lists forever.

PRESENTED BY AWS SUMMIT NYC

📅 Step into AWS Summit NYC on June 17

The Rundown: Connect, learn, and dive deep with AWS experts, developers, and industry leaders shaping the future of cloud and AI.

Explore:

• 200+ technical sessions and demos

• Keynote from VP, Agentic AI, Dr. Swami Sivasubramanian

• Interactive activations like 8-bit Agents Activation

• Real-world customer stories

Register here.

OPENAI

🚀 OpenAI’s ‘superapp’ revamp is coming soon

Image source: OpenAI / iSpot

The Rundown: OpenAI is reportedly planning its biggest ChatGPT revamp yet “in the coming weeks”, which will rebuild ChatGPT into an agent-and-coding "superapp" that prioritizes pushing its nearly 1B users toward paid products ahead of its upcoming IPO.

The details:

• The reorganization will place greater focus on its agentic coding platform, Codex, with a senior OAI member telling the Financial Times that “Chat is dead”.

• The move will combine coding, images, and third-party apps into one interface, believing model advances will now understand user intent across functions.

• The company also recently ran a new TV ad spot with the tagline “It’s time to fly”, pushing Codex and AI agents over simple ChatGPT-type interfaces.

• Codex has seen a 6x increase in users to over 5M since February, with OAI pushing to increase the number of paying business users as its IPO nears.

Why it matters: Claude Code was Anthropic’s breakout product, which was then simplified for its broader audience (albeit a bit awkwardly) with Cowork. OAI has now seen similar success and upgrades for Codex, and the “side quest” retreat is on full display via a superapp path that gets more paid users into its agentic coding product.

QUICK HITS

🛠️ Trending AI Tools

• 🎬 Frames.md - HeyGen’s format to create branded, consistent AI videos

• 🎆 Reve 2.0 - Reve's new 4K AI image model with new layout-based editing

• 🤖 ChatGPT - OpenAI’s AI assistant, with new ‘Dreaming’ memory upgrade

• 🌌 Ideogram 4.0 - Ideogram's new top-rated open-weight image model

📰 Everything else in AI today

Microsoft Azure's Copilot Migration Agent turns complex migration data into clear answers using natural language prompts. Read the whitepaper.*

Anthropic’s Mythos model started showing up in Dev Mode and across social media, with speculation of a public release coming this week.

Apple held a “top secret” meeting in early 2025 on the company’s AI crisis that led to Tim Cook taking the overhaul more seriously and becoming more involved in it.

Anthropic poached Clive Chen from OpenAI, who was an early hire on the company’s chip design team, coming amid rumors of Anthropic considering its own chips.

OpenAI introduced Lockdown Mode, a ChatGPT setting that disables features like live browsing, agent mode, and deep research to protect from prompt-injection attacks.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Laura S. in Dallas, TX:

"I built an automated workflow that creates personalized talking-dog videos as gifts. I love dogs and wanted a fun project to experiment with vibe coding. I asked friends and family for photos of their dogs and descriptions of their personalities. Then, I used AI tools to generate the script, voice, visuals, and final video.

Today, the workflow handles almost everything automatically, with just two human approval checkpoints: organizing photos, creating a recognizable source image, drafting the dog’s dialogue, generating the talking-dog video, polishing the edit, and delivering the finished product. The workflow is powered by Codex, ChatGPT, HeyGen, and Remotion working together behind the scenes.

What started as a fun experiment is now a repeatable workflow that can produce custom videos in under 30 minutes. It has become a creative way to combine my love of dogs, AI, and automation into personalized gifts that make people smile."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Anthropic confronts the RSI clock

• Read our last Tech newsletter: Google turns search into creator hubs

• Read our last Robotics newsletter: Sam Altman’s stealth robotics bet

• Today’s AI tool guide: Find extra prospects with this agentic framework

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Recursive self-improvement used to sound like a far-off sci-fi phrase. Now, frontier labs say the signs are showing up internally faster than anyone expected.

Anthropic's new report shows AI already accelerating the work that builds its own successors, and lays out the risks of systems that improve themselves… While also floating the jarring scenario of a coordinated development pause across the industry.

In today’s AI rundown:

• Anthropic charts path to self-improving AI

• OpenAI’s memory overhaul lets ChatGPT ‘dream’

• Stress test business ideas with Perplexity

• Rival AI labs unite behind bioweapons risks

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🔁 Anthropic charts path to self-improving AI

Image source: Anthropic

The Rundown: Anthropic just published "When AI builds itself," a report on recursive self-improving (RSI) systems — citing internal data on Claude's coding takeover and cautioning that fully self-improving AI could arrive before institutions are ready.

The details:

• Anthropic noted that RSI is not here yet or even inevitable, but said that Claude is advancing AI development “faster than we thought”.

• More than 80% of Anthropic's merged code was Claude-authored as of May, with engineers pushing 8x as much code per day in Q2 2026 as in 2024.

• "Each new version of Claude could be built by the version before it, without human involvement," co-author Jack Clark wrote of where the trend leads.

• OpenAI flagged the same loop this week in its “Democratic Governance of Frontier AI” blueprint, pointing to RSI's first sparks in today's systems.

• Anthropic said it would slow or pause frontier AI if peer labs did too, and plans policy talks in the coming months to discuss research, systems, and scenarios.

Why it matters: Anthropic and OpenAI aren't alone in feeling the RSI, with labs like MiniMax saying its M2.7 model helped build itself and new startups dedicated to the self-improvement loop popping up everywhere. The unknowns of RSI are scary, but it's also hard to fathom a feasible pause scenario that hinges on global coordination.

TOGETHER WITH LIGHTFIELD

💨 The fastest-growing AI-native CRM

The Rundown: Lightfield is the CRM used by over 4,000 companies that updates itself and does the work for you. Teach it how you sell with Skills, and watch it go to work for you.

Ask it anything, like:

• "Write follow-up emails to everybody I spoke with.”

• "Tell me why do we keep losing to our biggest competitor."

• “Find companies that look like my best customers and reach out to them.”

Try it for free at lightfield.app.

OPENAI

😴 OpenAI’s memory overhaul lets ChatGPT ‘dream’

Image source: OpenAI

The Rundown: OpenAI just introduced a new memory update within ChatGPT centered around "dreaming," a background system that turns past chats into a running, category-sorted profile of who you are for better personalization and evolving context.

The details:

• ChatGPT now keeps a running written summary of each user, grouped into areas like travel, hobbies, and work, replacing the previous list of one-off facts.

• Users can review memories, make corrections, add details, or ask ChatGPT not to bring up certain topics, with memory automatically updating over time.

• OpenAI says factual recall rose from 41.5% to 82.8% in its evals with dreaming, while preference-following climbed from 31.4% to 71.3%.

• Dreaming is rolling out to Plus and Pro users in the U.S., with Free and Go, and more countries getting the upgrade over the next several weeks.

Why it matters: Memory is one of the stickiest AI features in theory, but it’s been long overdue for an upgrade. Sam Altman frequently talks about the future of hyper-personalized AI, and dreaming may become a big part of forming that continuity and proactivity for users — and keeping them from switching to rivals in the process.

AI TRAINING

🤔 Stress test business ideas with Perplexity

The Rundown: In this guide, you will learn how to use Perplexity Deep Research to stress test any business idea. Save the prompt below once and rerun it on every idea you have to see what’s feasible to build.

Step-by-step:

1. Open Perplexity and switch to Deep Research mode. This works on the free plan (5 queries/day) and is basically a hidden version of Perplexity Computer

2. Paste this prompt with your idea in the chat, hit run, and walk away for 5 to 6 minutes. Perplexity does the research and builds the slide deck in the same run

3. Save the prompt somewhere you will actually use it again, like in a dedicated Perplexity space

4. Then, every Saturday morning, take one idea off your list and run it. You will burn through a year of half-evaluated ideas in a month

Pro tip: Build variants. A 6-slide version for a co-founder pitch, a version that compares two ideas, or a 90-day MVP plan for ideas that already cleared validation.

PRESENTED BY GITLAB

📡 GitLab Transcend goes live

The Rundown: GitLab Transcends is a live event streaming from London June 10-11, with an agenda built for developers – featuring live demos, real-world agentic AI use cases, and developer-first sessions.

With GitLab Transcends, you’ll get:

• Live demos of the Duo Agent Platform

• Agentic AI use cases from your peers

• The Developer Show, hosted by Snerio Developer Advocate, Colleen Lake

Come see intelligent orchestration, now with context. Register free.

AI & SAFETY

🧬 Rival AI labs unite behind bioweapons risks

Image source: Open letter to Congress

The Rundown: CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft just signed an open letter pressing Congress to make synthetic-DNA sellers vet every buyer/order, warning that AI can now enable bad actors to create bioweapon designs.

The details:

• Signees include Sam Altman, Dario Amodei, Mustafa Suleyman, Alexandr Wang, and Demis Hassabis, alongside DNA-synthesis industry leaders.

• The letter said, “AI systems now outperform PhD-level virologists… about highly technical lab procedures in their own domains of expertise.”

• The signers urge Congress to make U.S. synthetic-DNA and RNA sellers screen orders, verify buyers, and log sales to keep dangerous sequences traceable.

• They also warned that "knowledge barriers which have historically prevented bad actors from obtaining biological weapons will meaningfully erode."

Why it matters: Like RSI, biological weapons concerns have been outlined for years as one of the steps up the ladder of AI improvement — and this one’s concerning enough to even unite Altman and Amodei. It’s clear that laws and regulations are going to have to adapt quickly to a changing AI-accelerated world, but the question is if they can.

QUICK HITS

🛠️ Trending AI Tools

• 🌌 Reve 2.0 - 4K image model with new layout-based editing

• 🗣️ Miso One - Open TTS model that reads tones for expressive responses

• 🧾 Stack - Ramp's AI accounting OS for monthly bookkeeping

• 🧠 Nemotron 3 Ultra - Nvidia’s open 550B reasoning model for agents

📰 Everything else in AI today

The U.S. and Japan announced a $1B AI research partnership, becoming the first country to join the U.S. Genesis Mission’s push to double science output using AI.

Nvidia released Nemotron 3 Ultra, a fully open 550B reasoning model that runs 5x faster and up to 30% cheaper for agents, with performance similar to top open rivals.

Canadian Prime Minister Mark Carney introduced AI for All, a new five-year national strategy surrounding the tech that targets $200B in growth and 250,000 AI jobs.

Gopuff launched Go, an AI shopping assistant built with SpaceXAI that predicts users’ carts using Grok, real-time signals from X, and order data.

CloudFlare co-founder Matthew Prince revealed that bot traffic on the internet has now surpassed humans, with the milestone coming a year ahead of his expectations.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Anonoymous:

"I have 2 small kids at toddler age, and at times that can be very, very challenging. I have a scheduled task to run a pre-set prompt a couple of times a week to provide tips and tricks that can help me manage the day-to-day.

Sometimes it suggests an activity that I can try on the weekend, other times it suggests how to handle poor behaviour, time-outs, or how to handle bedtime routine. In sum, it helps expand the options available to me on how to deal with lots of different sides of parenting, helping me become a better father."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Reve, Ideogram trade prompts for layouts

• Read our last Tech newsletter: Microsoft’s first AI-native laptop

• Read our last Robotics newsletter: Sam Altman’s stealth robotics bet

• Today’s AI tool guide: Stress test business ideas with Perplexity

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. How you use an AI image model is starting to change, and two back-to-back releases show where it's going.

The new (open-source!) Ideogram 4.0 and Reve 2.0 make a similar case: the prompt gets you close, but letting users edit and control typography, regions, and layout after the fact is where the next breakthrough lives.

In today’s AI rundown:

• New image models swap text prompts for layouts

• Meta turns business chats into AI agents

• Automate social content calendar with Manus

• Study: AI tutors edge out law faculty

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

IDEOGRAM & REVE

🖼️ New image models swap text prompts for layouts

Image source: Ideogram / Reve

The Rundown: Two image labs just shipped new models, with Ideogram open-sourcing Ideogram 4.0 and Reve launching Reve 2.0 — and both pushing a more layout- focused, agentic iteration process for more user input and creative control.

The details:

• Ideogram 4.0 takes the top spot for open models and ranks behind only OpenAI and Google’s closed models on Design Arena.

• 4.0 excels at text rendering, typography, and graphic design, preferred by professional human designers over top rivals on Contra’s testing.

• Reve 2.0 surpasses Nano Banana 2 on Arena’s Text-to-Image leaderboard to take the No. 2 overall spot, trailing only GPT-image-2.

• The model’s outputs include labeled segments, allowing users to tweak specific parts of an image rather than regenerate it entirely.

• Reve creates the image ‘like code’, editing it by rewriting the layout rather than the prompt, while Ideogram uses a similar technique via JSON.

Why it matters: Image models have come a long way from the slot-machine days, when the only move was to re-roll the prompt. The real step change now is the granular control and editing that people used to jump to other apps for. With Ideogram, the open weights are the story: proving open source is not far behind the frontier.

TOGETHER WITH FUNCTION HEALTH

🩺 Your body has receipts

The Rundown: Function gives you access to 160+ lab tests, personalized insights, and ongoing tracking in one place — so you can see what's happening inside your body before symptoms show up.

A Function membership gives you:

• 160+ lab tests run annually across key health markers

• Private AI Chat trained on your individual results

• Clinician-reviewed insights to put your data in context

• Continuous tracking so you can see what changes over time

Own your health and use code RUNDOWN50 for $50 off your membership.

META

💬 Meta turns business chats into AI agents

Image source: Meta

The Rundown: Meta just launched Meta Business Agent globally, rolling out AI agents across WhatsApp, Instagram, and Messenger that can answer questions, book appointments, and close sales for any business on the platform.

The details:

• Meta first floated the feature in October as a free test in international markets, with over 1M businesses already utilizing the tool.

• Agents can close sales, recommend items, qualify leads, and book appointments across languages, with human takeover available at any point.

• A standalone Business Agent Platform plugs agents into outside tools like Zendesk and Shopify, with broader business operations abilities coming later.

• The agent is free to start, though Meta said it will be rolling out paid subscription tiers across different business sizes.

Why it matters: Meta’s sprawling social media empire is already a major driver for businesses, so embedding agents for them is a sharp move. The question will be trust. Stories like hackers fooling Meta's own support bot this week won’t help, as the tech giant hopes businesses will trust its agents with their own customers.

AI TRAINING

🤳 Automate social content calendar with Manus

The Rundown: In this guide, you will learn how to build a custom content calendar in Manus that updates, plans, and generates each week of social posts for you. It will automatically save generated posts in your Google Drive so you can easily access them.

Step-by-step:

1. Sign up for Manus and get its desktop app, then create a Google Drive folder and add your brand documents, briefs, example posts, or product notes

2. Open the Manus Connectors page, choose Google Drive, authorize your account, then create a new project for this calendar

3. Prompt: “Build a one-page HTML content calendar I can use as a visual planning dashboard for one week of content. Use the brand documents in my Google Drive folder to generate post titles, captions, and assets for [Instagram, LinkedIn, X, email]. Show each post and asset in the calendar page, and save the generated posts and assets back into Google Drive”

4. Tell Manus to generate the first week of posts based on the brand assets you added to Google Drive, then turn that planning process into a reusable skill

Pro tip: Once the skill is working, ask Manus to make it a weekly automation that refreshes the next seven days of posts and updates the calendar.

PRESENTED BY VANTA

👋 Find your Calm-pliance

The Rundown: Your employee just added their 67th AI tool. Congrats. You also just created your 67th security blind spot. The Vanta Agent is the sharpest GRC engineer you’ve never had to hire, working tirelessly across the platform to draft policies, complete questionnaires, and flag issues before they escalate.

Here's what Vanta Agent takes off your plate:

• Continuous policy drafting as your tool stack grows

• Security questionnaires completed automatically

• Compliance gaps flagged before they become audit findings

• Always-on coverage across your existing Vanta workflows

Don't be the last one scrambling before an audit — watch the on-demand demo to learn more.

AI RESEARCH

⚖️ Study: AI tutors edge out law faculty

Image source: Images 2.0 / The Rundown

The Rundown: A new study led by Stanford ran a blind test of AI legal tutoring, asking 16 contract professors to judge anonymized answers from their peers and from two of Google's AI systems — with the faculty siding with the AI outputs 75% of the time.

The details:

• The study tested contract-law office-hours questions, a setting where strong answers need judgment and critical thinking instead of just one correct answer.

• Sixteen professors from 14 schools blindly judged 2,918 matchups between their own answers and those of Google’s Gemini 2.5 Pro and NotebookLM.

• Faculty chose Gemini 2.5 Pro and NotebookLM responses 75% of the time, with only a single top professor staying level with the models in evaluations.

• Extending the testing with an AI stand-in judge, the team ranked nine more systems, with Claude Opus 4.7 on top and all models beating the professors.

Why it matters: Early models like GPT-4 were already passing the bar exam, but this study puts AI in tougher, more subjective judgment situations on office hours contract law questions. AI’s rollout into the education world is still controversial and jagged, but areas like on-demand tutoring can help change the learning process for the better.

QUICK HITS

🛠️ Trending AI Tools

• ⚡ CData Connect AI - Give Claude, ChatGPT, Copilot or any AI tool live, governed, read & write access to your business data in one unified layer.*

• 🎆 Ideogram 4.0 - Ideogram's new top-rated open-weight image model

• 👋 Meta Enterprise Agent - Meta’s AI agent for customer sales and support

• 💻 Personal Computer - Perplexity’s local orchestrator, now on Windows

*Sponsored Listing

📰 Everything else in AI today

Suno raised over $400M at a $5.4B valuation, with the AI music startup rolling out its first model built in partnership with the music industry “in the coming months.”

Google released Gemma 4 12B, a new multimodal model able to run on a 16GB laptop, and the first Gemma variant of this size built for native audio.

xAI rolled out Grok Imagine 1.5 Preview, the company’s latest image-to-video update, which brings upgraded realism, audio syncing, and prompt following.

Microsoft and the Mayo Clinic are building a frontier healthcare AI, trained on anonymized patient data and owned by the clinic, with Azure handling distribution.

Google Labs launched a new experiment called Dreambeans, connecting to users’ Gmail, Calendar, Photos, and Search to generate a feed of personalized daily stories.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Jeff M. in Great Falls:

“A simple but surprisingly useful use of AI came while shopping for a used vehicle. I had a browser page open with hundreds of listings and asked Gemini in Chrome to extract the data and plot it by price, model year, and mileage. Within seconds, it generated a visualization showing the distribution of asking prices across different years and mileage ranges.

The chart made it easy to see how much value was gained or lost by moving one model year newer or older, and quickly highlighted listings that deserved further investigation. It also exposed suspicious outliers, such as unusually low-priced vehicles that appeared to be stale or bait listings designed to generate leads and encourage buyers to contact the dealership.

What would have taken hours became digestible in a matter of seconds.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Microsoft paves its own AI way at Build

• Read our last Tech newsletter: Microsoft’s first AI-native laptop

• Read our last Robotics newsletter: Nvidia’s plug-and-play humanoid

• Today’s AI tool guide: Automate social content calendar with Manus

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. For years, Microsoft's AI story was really OpenAI's story. At Build 2026, it felt like the tech giant was finally writing its own path.

With seven new in-house models, an OpenClaw-based agent, agent-first hardware, and more, Microsoft stopped looking like a distribution partner and started looking like the company rebuilding the agent era around its own rails.

In today’s AI rundown:

• Microsoft kicks off Build with models, agents, and qubits

• Trump softens AI executive order into voluntary review

• Use Claude Design’s slide decks feature like a pro

• Legendary Hollywood director opens the door to AI

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

MICROSOFT

🏗️ Microsoft kicks off Build with models, agents, and qubits

Image source: Microsoft

The Rundown: Microsoft just made a full-stack agentic AI push at Build 2026, releasing new models, its first always-on agent, a quantum chip, a platform for “agent-first” devices — positioning Windows and Microsoft 365 as the control layer for agents.

The details:

• Microsoft AI released seven new MAI in-house models spanning reasoning, coding, image, voice, and transcription, with access through Microsoft Foundry.

• Microsoft Scout, its first "Autopilot" agent built on OpenClaw, runs in Teams and takes proactive actions like scheduling meetings and prepping materials.

• Majorana 2, a quantum chip AI agents helped design, shows a 1,000x reliability improvement, speeding timelines for a usable machine to as soon as 2029.

• The company previewed Project Solara, an upcoming platform for agentic devices, with hardware concepts including a badge and desk companion.

• On the heels of its Surface Laptop Ultra reveal, Microsoft also introduced the Surface RTX Spark Dev Box, a new mini-PC built for AI workloads.

Why it matters: Build is a timely piggyback off of Nvidia’s big day, with Microsoft following the agentic theme and even partnering with the chipmaker for AI laptops and PCs. With in-house models, an OpenClaw agent, and agentic hardware, Microsoft is paving a bold new path in the first year of independence from OpenAI’s shadow.

TOGETHER WITH TELEPORT

🚫 Don’t give AI agents standing access

The Rundown: AI agents are touching your production systems without an identity. Teleport gives every autonomous agent a unique, cryptographic identity with least-privileged, JIT access that expires automatically.

With Teleport, you get:

• 4.5x fewer security incidents

• An audit log of every prompt, query, and tool call

• A unique cryptographic identity per agent

Ready to Teleport?

AI & THE GOVERNMENT

⚖️ Trump softens AI executive order into voluntary review

Image source: The White House

The Rundown: U.S. President Donald Trump signed an executive order asking AI labs to voluntarily hand frontier models to the government for a 30-day security review before release — retreating from the previously expected 90-day requirement.

The details:

• Trump scrapped a 90-day draft hours before a planned May 21 ceremony, telling reporters it would "get in the way of" the U.S. AI race with China.

• Labs are asked to share “covered frontier models” flagged by a classified process as capable of finding security flaws in a 30-day window before launch.

• Former AI czar David Sacks, who reportedly fought the first draft, came around once the review window shrank from 90 days to 30.

• The order also rules out any mandatory licensing or permits for new models, and directs the DOJ to go after AI-powered hacking of computer systems.

Why it matters: With Anthropic’s Claude Mythos nearing a public release and GPT-5.6 likely to have similar capabilities, the cybersecurity threats are becoming very real, very fast. But this EO looks more like the government pushing for a (voluntary) front-row seat to the frontier AI action more than a fix for a potential security concern.

AI TRAINING

👨‍💻 Use Claude Design’s slide decks feature like a pro

The Rundown: In this guide, you will learn how to use Claude Design to turn your raw data into a useful strategy deck complete with actual insights. Claude Design analyzes what is working and gives concrete recommendations your team can use.

Step-by-step:

1. Start with a messy CSV report (YT data, Facebook ads, etc.) and decide what the deck needs to do (like find patterns and turn it into a repeatable system)

2. Open claude.ai/design, choose Slide deck, skip the design system, toggle on speaker notes, and upload your data

3. Prompt: “Turn these files into a strategy deck on performance. Analyze the results by item and extract best practices from the data and assets. Use charts, rankings, and concrete recommendations. Match images or creative files to CSV using the filename or matching field. Keep it presentation-ready”

4. Generation will take 10-15 minutes. You can export to PowerPoint or Google Slides when it's ready

Pro tip: Duplicate the project and upload more data sources for Claude to incorporate into the presentation.

PRESENTED BY GITLAB

🚀 The future of agentic AI dev

The Rundown: Agentic AI is reshaping software development, and the leaders adopting it early are setting the pace. GitLab Transcend streams live from London on June 10 with unfiltered insights from peers deploying it now — plus an early look at where GitLab is taking it next.

With GitLab Transcend, you'll get:

• Unfiltered insights from peers deploying agentic AI

• An early look at GitLab's roadmap

• New research and live demos

• All virtual, streaming from London

Register for free today.

AI & HOLLYWOOD

🎬 Legendary Hollywood director opens the door to AI

Image source: Black Forest Labs

The Rundown: Martin Scorsese just went public as an adviser to AI image startup Black Forest Labs, with a new video detailing the Academy Award-winning director using the company’s FLUX model to help storyboard a new film.

The details:

• Scorsese signed on last year as a partner and adviser to Black Forest Labs, utilizing its FLUX AI image models for preproduction of a new film.

• His use is for storyboarding only, with no generated actors, sets, or footage — calling being able to share a storyboard instantly "creatively freeing."

• Scorsese also said that “Cinema is a young medium, only around 125 years old, so we have to be open to how it can evolve”.

Why it matters: This is going to ruffle some anti-AI Hollywood feathers, but a figure of Scorsese’s level coming out in support of an AI tool is a big deal. It’s also a good example of using the tech in a way that aids the creative process without replacing it, a more stomachable entry point for skeptical filmmakers than full AI films or studios.

QUICK HITS

🛠️ Trending AI Tools

• 🤖 MAI - Microsoft AI’s in-house model family, with seven new models

• 🖥️ Holo3.1 - H Company’s upgraded computer use model, running fully locally

• 💻 Hermes Desktop - Nous Research's agent as a native desktop app

• 🚀 Codex - OpenAI’s agentic coding tool with new Sites, role-specific plugins

📰 Everything else in AI today

Anthropic expanded Project Glasswing, giving 150 new organizations across 15 countries access to its powerful Claude Mythos Preview model.

OpenAI launched new Codex updates, including Sites that allow users to create and share hosted websites and apps, alongside new role-specific plugins across domains.

Cognition rebranded its Windsurf IDE as Devin Desktop, a single surface for running agents locally or in the cloud, with support across agents like Codex and Claude.

Famed investor Elad Gil said humanity is “likely in very early lift off and exponential”, saying the models released in December crossed a major technological threshold.

Google's parent company, Alphabet, is selling $80B in stock to fund its AI infrastructure expansion, with the company planning up to $190B in spending this year.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader An in Cape Town, Africa:

"I wanted to analyse the income and expenses of our Airbnb properties to see trends and if Claude could give me some recommendations. So I used Cowork and pointed it to my folder where all income and expenses info (including the agent, property management costs, and income, rates and taxes, internet costs, etc.) is stored.

Claude analysed all the files and came up with a Five Tab Dashboard: Overview, Net Income & Trends, Expenses, Occupancy & ADR, and Recommendations. It is fully interactive, and I can look at each property individually and all together, see trends over the years, and look at how low season impacts our income.

I used to have to go through every single invoice and put it into a spreadsheet, which took a lot of time, and therefore, I did not do it as often as I should or wanted to. Now, with all the graphs and everything included, it makes it much easier to analyse the data.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Nvidia corners the AI agent stack

• Read our last Tech newsletter: Microsoft’s first AI-native laptop

• Read our last Robotics newsletter: Nvidia’s plug-and-play humanoid

• Today’s AI tool guide: Use Claude Design’s slide decks feature like a pro

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Nvidia CEO Jensen Huang had a simple but bold statement at COMPUTEX in Taiwan to open his keynote: “Agentic AI has arrived.” He then pointed his whole ecosystem at backing that claim up.

The chipmaker's latest reveals stretched from the laptop to the data center to the factory floor, unified by the bet that AI agents will dominate future compute demand far more than the people using them.

In today’s AI rundown:

• Nvidia threads agents across the stack

• Bernie Sanders seeks a public AI stake with new bill

• Turn Claude sessions into skills with a daily audit

• Hackers access IG accounts by… asking Meta AI?

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

NVIDIA

💻 Nvidia threads agents across the stack

Image source: Nvidia

The Rundown: Nvidia just introduced a series of new AI releases across hardware, robotics, models, and more at COMPUTEX 2026, all built around the idea that agents will soon be the biggest consumers of compute power.

The details:

• The new RTX Spark supercomputer chips built with Microsoft run AI agents directly on PCs, with Nvidia saying it takes Windows “from tool to teammate.”

• Nvidia called Vera the “CPU for agents”, a processor that finishes tasks 1.8x faster than rivals and is now being used by Anthropic, OpenAI, and the NYSE.

• Cosmos 3 is a new open robotics model, giving robots and self-driving cars the ability to plan ahead and anticipate moves instead of just reacting.

• Nemotron 3 Ultra is a new 550B parameter model that moves to the top of U.S. open-source and competes with Chinese rivals like Qwen3.5 and Kimi K2.6.

Why it matters: Nobody builds across the entire tech stack quite like Nvidia, but the central theme in the chipmaking giant's latest moves is the prioritization of AI agents themselves as the consumers of compute — with a company worth $5T+ now organizing its entire lineup around enabling software that didn't exist two years ago.

TOGETHER WITH YOU.COM

🔍 What makes a good web search API?

The Rundown: If you’re looking for hallucinations and unpredictable failures, pick a search provider by running a few test queries and hoping for the best. If you’re looking for accurate, reliable, fresh data, this technical guide from You.com gives you access to an exact framework to evaluate AI search and retrieval.

What you’ll get:

• A four-phase framework for evaluating AI search

• How to build a golden set of queries that predicts real-world performance

• Metrics and code for measuring accuracy

Go from “looks good” to proven quality. Learn how to run an eval.

AI & THE U.S. GOVERNMENT

🏛️ Bernie Sanders seeks a public AI stake with new bill

Image source: Erik Carter / The New York Times

The Rundown: In an NYT op-ed, Bernie Sanders previewed the American A.I. Sovereign Wealth Fund Act, an upcoming bill proposal that would route half the stock of the largest AI companies into a public fund and pay the gains back to Americans.

The details:

• Sanders framed it as a one-time tax collected in equity, with the government gaining voting power and a board seat at OAI, Anthropic, and xAI.

• Sanders cites the AI labs themselves as precedent, with each already pitching public funds or "universal high income" to spread AI's gains.

• He pointed to Norway’s $2T oil fund and Alaska’s oil dividends to residents as examples of “ensuring ordinary people benefit from national wealth”.

• Sanders said, “A.I. is being built on a public resource far more valuable than oil: the accumulated knowledge, creativity, and labor of mankind.”

Why it matters: With public AI sentiment at a low and mega IPOs from top AI labs coming at valuations that leave little room for the average investor, Sanders’ concerns that ordinary people are not benefiting are real. Getting the AI labs on board to give up 50% of their equity, despite the posturing, is another question.

AI TRAINING

✍️ Turn Claude sessions into better skills with a daily audit

The Rundown: In this guide, you will learn how to make Claude Cowork look through dozens of session files and improve the skills you use. It will also suggest new skills and help you turn the audit into a repeatable skill and automation.

Step-by-step:

1. Open Cowork, start a chat, and prompt to scan sessions from the past 7 days and suggest with evidence if any personal skills should be created or improved

2. Review the suggestions and approve the ones with repeated friction, repeated corrections, or a workflow you will reuse. Reject anything that feels one-off

3. Tell Claude to turn the audit process itself into a skill, and then tell it to set up an automation to run the audit weekly or daily

4. You can open the "Scheduled Task" tab to adjust timing or move it to a lower-cost model like Sonnet

Going further: Run the same audit on automations. Ask which scheduled tasks are noisy, stale, expensive, or missing approval steps, then improve the ones with evidence.

PRESENTED BY UNWRAP

⚡ See how Oura automates customer feedback analysis

The Rundown: Oura was sitting on thousands of signals (surveys, app reviews, support tickets, social comments) with no way to connect them fast enough to act. Issues hid in the noise until they became problems. With Unwrap, they turned that fragmented data into a single source of truth. Analysis that took days or weeks now happens instantly.

Unwrap's customer intelligence platform gives you:

• All feedback automatically categorized across every channel

• Alerts that surface the most actionable insights straight to the right owner

• Natural language queries via Assistant or your favorite tools using Unwrap’s MCP

Unwrap is offering free trials to Rundown AI readers. Grab 15 minutes with the team to get set up.

META & AI SECURITY

🔓 Hackers access IG accounts by… asking Meta AI?

Image source: 404 Media

The Rundown: Meta just fixed an Instagram security flaw that allowed hackers to use its AI help tool to take over prominent profiles by simply asking the chatbot to use a different email address and gaining the codes needed to access the accounts.

The details:

• Meta gave its AI the power to handle password resets on Facebook and Instagram starting in March, with the exploit reportedly active for months.

• Accounts hacked include a dormant Barack Obama account, Sephora, and Space Force head John Bentivegna, with accounts then resold in minutes.

• Hackers switched a VPN near a target’s region and asked AI support for a password reset and email change, and AI sent a code to the new email.

• Meta told 404 Media that the exploit “has been resolved and we are securing impacted accounts.”

Why it matters: This isn’t the first AI support story gone wrong, and won’t be the last — but the bigger issue might be the reach of a company like Meta (which has frontier AI ambitions) turning its entire support process over to a tool so easily exploited. Hackers might not need Mythos-level capabilities… Sometimes you just have to literally ask.

QUICK HITS

🛠️ Trending AI Tools

• 💡 AhaCreator 3.0 - Your 24/7 AI agent for influencer marketing*

• 🚀 M3 - Minimax’s new open-weight model with 1M context and computer use

• 🎥 Aleph 2.0 - Runway’s SOTA video editing model

• 🤖 Claude Opus 4.8 - Anthropic’s newly updated top model class

*Sponsored Listing

📰 Everything else in AI today

Anthropic announced it has filed with the U.S. Securities and Exchange Commission to go public, officially pushing forward on the race with OpenAI to hit the open markets.

Apollo’s chief economist shared data claiming “zero evidence of job losses because of AI”, saying that “cheaper technology is creating more demand and more jobs.”

MiniMax released M3, an open-weight model it claims tops GPT-5.5 and Gemini 3.1 Pro on coding benchmarks while approaching Anthropic's Opus 4.7.

OpenAI started construction on ‘The Barn’, a 1 GW Stargate data center campus in Michigan, promising 2,500 union jobs and $45M in Codex credits for in-state students.

Florida’s AG brought the first state-led lawsuit against OpenAI and CEO Sam Altman, alleging ChatGPT played a role in planning mass shootings and instances of self-harm.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Johannes P. in Ilomantsi, Finland:

"I turned The Rundown into a personal AI idea bank. I saved many of your newsletters into one email folder, downloaded them to my computer, and asked an LLM to analyze the archive. I wanted to find the most useful ideas and AI use cases.

I also gave the model a simple “about me” markdown file, explaining what I do, what I care about, and what kinds of AI tools are useful for my work and everyday life. Then I asked it to rank the Rundown examples based on what would be relevant to me.

This made the newsletter much more useful. Instead of reading each issue once and forgetting many good ideas, I can now ask: “Which of these examples should I try?” or “What use cases are most useful for my research, coding, and writing?”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: AI’s next dataset is your apartment

• Read our last Tech newsletter: Meta launches paid tiers across its apps

• Read our last Robotics newsletter: Nvidia’s plug-and-play humanoid

• Today’s AI tool guide: Turn Claude sessions into better skills w/ daily audits

• RSVP to our next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. A startup will clean your New York apartment for free. The bill comes due in a currency you might not expect: the data captured during the process.

MicroAGI’s new Shift app subsidizes the cleanings by recording workers on the job, turning a free service into a strange preview of AI’s next data hunt — where humans are the customer, the labor, and the training material for automation.

In today’s AI rundown:

• Startup cleans apartments in exchange for AI data

• The Rundown Roundtable: Our AI use cases

• Build a video workstation with Higgsfield and Claude

• Ex-DeepMind group tackles self-improving science AI

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

PHYSICAL AI

🧹 Startup cleans apartments in exchange for AI data

Image source: Shift

The Rundown: German startup MicroAGI’s Shift app just opened a free home-cleaning service in New York City that records its cleaners through head-mounted cameras, trading chores for first-person data to both sell to AI labs and use in its own AI research.

The details:

• A vetted cleaner shows up wearing a camera that co-founder Bercan Kilic calls a "magic hat," filming the roughly two-hour job point-of-view style.

• Despite covering the cost of the cleaning, the human footage is worth more to robot makers for training, letting Shift cover the bill and still profit.

• Shift's site claims to already pay people across the world $20 an hour to film everyday chores, with $5M+ paid out in Q1 across a variety of tasks.

• GM Harry Kilberg said the launch drew "thousands and thousands of bookings," with New York first and London, Munich, and Zurich next.

Why it matters: As we've seen with DoorDash paying couriers to capture task data, the next AI dataset is coming from ordinary human work instead of the internet. Shift pushes that model deeper into the home, where people are both the customers getting free service and the workforce teaching robots how to replace pieces of the job.

TOGETHER WITH NEBIUS

🚀 Open-source LLMs in real production

The Rundown: Nebius Token Factory is an AI cloud platform that lets you run open-source LLMs in real production. Capture live traffic, fine-tune and optimize, then deploy your own checkpoints to dedicated GPU endpoints, with stable latency, predictable cost, and clear data residency.

With Nebius, you can:

• Deploy 60+ open-source models via one API

• Fine-tune and serve your own checkpoints

• Scale to 100M+ tokens per minute

Go from LLM to production system on one platform.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

Image source: Ideogram / The Rundown

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Zach, AI Writer: It’s finally gardening season in the Midwest, and I wanted to do a big overhaul of my space with a new large in-ground bed and a more organized layout of crops.

I first took some images of the space and had ChatGPT / Codex generate some visualizations of what it could look like based on my rambling list of requirements and measurements. After deciding on the vision, I asked for help creating a materials list, a step-by-step guide to building the bed (with instructions on sawing, drilling, etc.).

I also provided a list of what we wanted to grow, and was able to get a well-planned map of where to place each plant, optimizing for aspects like shade vs. sun, companion plants, vining, etc. Several trips to Home Depot and a couple of days of work later, we have an awesome, refreshed space and a thoughtful, organized garden for the season!

Joey, Partnerships: I recently had an Evolt body scan. I've uploaded my results to Claude, shared my daily habits (sleep, diet, schedules, etc.), and asked it to create a full workout and diet plan based on the areas I want to improve.

I was also able to include my recent Oura ring report to see if there's anything wrong with me. Will I follow it? We'll see....

AI TRAINING

📽️ Build a video workstation with Higgsfield and Claude

The Rundown: In this guide, you’ll learn how to build a short-form video workstation with Higgsfield + Claude Code that can generate campaigns, track results, improve prompts from feedback, and turn successful workflows into reusable skills.

Step-by-step:

1. Create a dedicated project folder, open it in Claude Code, install the Higgsfield CLI, authenticate your account, and add the Higgsfield skills package

2. Ask Claude to interview you about brand, audience, and content goals, then set up campaign folders (with input, output, draft, and final folders), files for brand guidelines, tracking, and a workflow README

3. Create the first campaign by prompting to generate two video concepts, turning them into Higgsfield-ready prompts, then, once approved, generating the videos and saving them in the output folder

4. Finally, turn the campaign workflow into a reusable skill, then create a second skill that improves future videos from feedback, prompt revisions, and notes

Pro tip: Run at least a few campaigns manually before automating. The best skills come from real feedback.

PRESENTED BY GITLAB

🚀 The future of AI dev, live

The Rundown: GitLab Transcend is a free virtual event streaming live from London on June 10, with regional replays for APAC and AMER on June 11 — packed with live demos and agentic AI use cases from your peers.

With GitLab Transcend, you'll get:

• Live demos of Duo Agent Platform

• Agentic AI use cases from your peers

• The Developer Show with Colleen Lake

Register now, free and virtual.

INHERENT LABS

🧪 Ex-DeepMind group tackles self-improving science AI

Image source: Inherent Labs

The Rundown: Several Ex-Google DeepMind employees came out of stealth with $50M for Inherent Labs, a London startup building an AI science platform that puts scientists alongside self-improving AI to work out which problems are worth pursuing.

The details:

• Co-founders Tantum Collins, Edward Hughes, and Louis Kirsch came from DeepMind, while Kaloyan Aleksiev previously worked at Reka AI and Microsoft.

• The Faraday platform will pair researchers with self-improving agents built to spot higher-value scientific questions instead of only answering prompts.

• The lab says it will test recursive self-improvement across the research org, including everything from agent training to resource allocation and decisions.

• Inherent is also exploring what “AI taste” looks like in science as the research process shifts, and how humans and machines can best work together.

Why it matters: Self-improving AI is a quest many of the top AI labs and newly funded startups are trying to tackle, and Inherent adds another group with strong pedigrees to the list. Inherent is applying the recursive logic to science itself, making the entire lab and organization the loop instead of just the model training.

QUICK HITS

🛠️ Trending AI Tools

• ⚙️ Codex - OAI’s agentic coding tool, with new feature expansions on Windows

• 🚀 Claude Opus 4.8 - Anthropic’s newly updated top model class

• 🍎 Koji by Brilliant - AI tutor for math and coding

• 🫂 Dynamic Workflows - Deploy subagents within Claude Code

📰 Everything else in AI today

5-Day AI Agents Vibe Coding Intensive with Google, June 15–19 – Learn to build real AI agents in just five days. Registration closes tomorrow. Sign up now.*

Microsoft is reportedly set to merge GitHub Copilot, chat, Cowork, and Autopilot into one super app, matching similar all-in-one plays OAI and Musk's X are already making.

SoftBank committed up to $87B to build what will be the largest AI data center project in France, extending an infrastructure spree that includes a $60B+ OAI stake.

Nvidia unveiled RTX Spark, a superchip designed for personal AI agents, bringing up to 1 petaflop of AI performance to Windows laptops with all-day battery life.

Sam Altman posted that OpenAI is hiring for its Robotics division, saying that the company imagines “everyone having a personal robot doing anything they need.”

OpenAI launched Rosalind Biodefense, giving the U.S. government and vetted partners access to its biology AI for pandemic-preparedness and outbreak-response.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Wyatt B. in Athens, AL:

"I’m using a product within HubSpot (SalesHub) called Prospecting Agent. The prospecting agent can be set up to run specific “plays” where I enroll companies from our CRM (as well as contacts associated with those companies) and constantly look for buying signals and other trigger events that could provide a warm lead.

For example, if I enroll a fictions company called Acme Corp, and associated contacts John and Jane Smith, my prospecting agent will look for buying signals in relevant news, articles published by the company or contacts, acquisitions, stock alerts, or other relevant events that could warm a lead so my SDRs aren't going in cold.

I can also build custom sequences where the agent will take the trigger event and write an email from one of my SDRs or me and reach out to the prospect, and choose to approve on demand. This has allowed my company to hire fewer SDRs, allowing existing SDRs to focus on prospecting and closing deals faster."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Anthropic just eclipsed OpenAI

• Read our last Tech newsletter: Meta launches paid tiers across its apps

• Read our last Robotics newsletter: Figure’s humanoids get a retail job

• Today’s AI tool guide: Build a video workstation with Higgsfield and Claude

• RSVP to next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Anthropic just took the lead from OpenAI — not just with a more powerful frontier model, but in the market too.

With the new Opus 4.8 crushing on benchmarks, a more powerful Mythos on the way, and a valuation closing in on $1T, Dario and co finally seem to have all the pieces in place for a blockbuster public listing.

In today’s AI rundown:

• Anthropic’s Opus 4.8, nearly $1T valuation

• Apple’s new AI Siri to take on ChatGPT

• Use Codex to build a game in one prompt

• AI doubles dev output, but not for everyone

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🚀 Anthropic’s Opus 4.8, nearly $1T valuation

Image source: Anthropic

The Rundown: Anthropic made two big announcements on the same day — Claude Opus 4.8, which crushes almost all major benchmarks, and a massive funding round that makes it the most valuable AI lab in the world.

The details:

• Coming at the same price as 4.7, Opus 4.8 bests GPT-5.5 and Gemini 3.1 Pro on agentic coding, computer use, financial analysis, and Humanity’s Last Exam.

• 4.8 is the least lazy among all Anthropic models and more honest, with increased likelihood to flag uncertainties instead of making unverified claims.

• The model’s Fast mode is 3x cheaper, plus claude.ai gets new effort control, and Claude Code gets parallel sub-agents for complex, long-running tasks.

• Anthropic paired the release with a $65B raise, taking its valuation to $965B (beyond OpenAI), and the promise of a Mythos-class AI in “the coming weeks.”

Why it matters: While the race isn’t over, Anthropic has crossed a milestone that would have seemed unlikely two years ago — a valuation higher than OpenAI and a model that leads on nearly every benchmark. Clearly, its safety-first push is paying off commercially, even as Sam Altman calls the strategy “fear-based marketing.”

TOGETHER WITH UNWRAP

⚡ See how Oura automates customer feedback analysis

The Rundown: Unwrap’s customer intelligence platform that pulls all your feedback — surveys, reviews, support tickets, social comments — into one view, then uses AI to surface the most actionable insights and deliver them to your inbox. Teams at Perplexity, Stripe, Oura, lululemon, and DoorDash rely on Unwrap to ensure no customer voice gets lost.

With Unwrap, you get:

• All customer feedback automatically categorized

• Query feedback using Unwrap Assistant, or in your favorite tools using Unwrap’s MCP

• Real-time alerts from feedback as they arise

• A clear view of customer sentiment

Unwrap is offering free trials to Rundown AI subscribers! Grab 15 minutes with the team to get set up.

APPLE

🧠 Apple’s new AI Siri to take on ChatGPT

Image source: Bloomberg

The Rundown: Apple’s long-overdue AI Siri finally appears to be taking shape, with Bloomberg giving a glimpse of the revamped assistant, rebuilt on Google Gemini, with a dedicated ChatGPT-style app and support for third-party AI agents.

The details:

• Siri will live inside Dynamic Island, with a swipe-down interface to run AI search, chat, or iOS tasks, using on-device data, screen content, and the web.

• The assistant, rebuilt on Google Gemini, will run AI-powered web searches similar to Perplexity, with answers surfacing as rich cards in the Dynamic Island.

• Swiping down farther on the cards brings up a dedicated ChatGPT-style Siri app, with users getting the option to route queries to external AI models.

• The revamped Siri will land in the Camera app, with advanced AI-based photo editing, wallpapers, and natural language shortcut-creation also in the pipeline.

Why it matters: Apple has been very slow in the AI race, promising features in 2024 that never shipped, while OpenAI and Google pulled ahead. If this overhaul lands, Apple’s 1B+ iPhone users will experience AI through the phone they use every day. But if it doesn’t, there will be a lot on John Ternus’ plate as he takes over as the new CEO.

AI TRAINING

👾 Use Codex to build a game in one prompt

The Rundown: In this guide, you will learn how to use Codex /goal to build a small browser game without nudging the agent every few minutes. The game is the demo, but the real move is learning to give Codex a finish line it can work toward on its own.

Step-by-step:

1. Open Terminal and enable goals with: “codex features enable goals.” Then, think of a short, simple game idea with rules you can test

2. If the idea feels fuzzy, ask ChatGPT to rewrite it in 100 words or fewer with objective tests. If it cannot do that, the scope is probably too big for one /goal

3. Paste the description after /goal, then follow the checklist Codex creates. For a simple game, expect about 5–6 minutes of building, testing, and fixing

4. When testing, give feedback as another /goal command. Be specific with instructions, like: “Add distinct animations for every action the user can take”

Pro tip: This works beyond games. List three business processes, choose a metric that proves each works, then ask Codex to improve the process against that metric.

PRESENTED BY DATADOG

📊 Scale your LLM applications with confidence

The Rundown: Datadog’s free guide walks teams through building observability into their LLM stack, so engineers can move fast on production AI without flying blind on errors, costs, or security risks.

In the guide, you’ll learn how to:

• Monitor LLM workflows for errors, latency, and token costs

• Detect prompt injections and sensitive data exposure before they escalate

• Evaluate output quality at scale with built-in and custom checks

Download the LLM Observability Best Practices Guide for free.

AI RESEARCH

🧑‍💻 AI doubles dev output, but not for everyone

Image source: Cursor

The Rundown: Cursor released its Developer Habits Report, based on its own product and engineering data, showing dev output has more than doubled, but the gains are concentrated, with a small group of power users pulling far ahead of everyone else.

The details:

• Lines of code added by each dev per week have gone from 3.6K to 8.6K in 18 months, with mega PRs with 1K+ lines changed becoming more common.

• Agents are doing more end-to-end work, with tool calls up 30% in two months and 5x more AI-made changes reaching commits without manual review.

• Cost per agent request varies 9x across models (Opus 4.7 being the most expensive), meaning the cost of one workflow can vary a lot with underlying AI.

• However, the gains remain concentrated, with the top 1% of devs producing 46x more code than the median active user and the gap widening every month.

Why it matters: AI doing deeper work and contributing more code falls in line with growing capabilities, but the usage gaps are worth noting. Not everyone is capturing the full productivity gains, and with the cost per agent request varying heavily across models, many teams may not be using the most cost-efficient AI for their tasks.

QUICK HITS

🛠️ Trending AI Tools

• 💡 Founder Starter Kit - Pika’s Claude skills to go from product to launch

• 🎙️ Dubbing V2 - ElevenLabs’ new dubbing AI that adapts across 90 languages

• 📽️ Paris 2.0 - Bagel’s efficient, decentralized-trained video generation AI

• 💻️ Computer - Perplexity’s agent, now inside Excel, Word, and PowerPoint

📰 Everything else in AI today

Workday DevCon Digital Experience, June 2 & 4 — Build agents. Build skills. Build your career on Workday’s developer platform. Register today.*

Google doubled Omni generations for Ultra users and fixed Gemini’s usage-limit issues with free Flash-Lite prompts, caps on high-cost requests, and improved tracking.

Elon Musk said that SpaceX’s compute deal with Anthropic is for 180 days, not three years (as mentioned in the S-1 filing), but indicated a longer deal is possible.

CNN filed a lawsuit against Perplexity, alleging the startup’s AI tools generate a “verbatim” copy of its articles while providing information locked behind a paywall.

An AI consultant revealed to Axios that their client accidentally spent nearly $500M in one month after failing to set usage limits on employees’ Claude licenses.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Gabriela in Austin, TX:

“I really wanted to know what my best colors are. I uploaded several selfies to ChatGPT and asked it to perform a color palette analysis. The results aligned with my best guesses, so I moved on to hairstyles based on face shape and outfits based on my body style. I uploaded more photos and shared my measurements, and had it get really specific about colors, styles, shapes, etc.

Next, I asked ChatGPT to write those into custom instructions to create a Project within ChatGPT just for style. I used what it wrote, making minor revisions, then started the Project with some additional photos included (some of me, some for inspiration).

Now I use that Project to get feedback on daily outfits, haircuts, new clothes, etc. It’s been working great, and I feel more confident and more myself.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: A world model for proteins is here

• Read our last Tech newsletter: Ferrari’s first EV, designed by Jony Ive

• Read our last Robotics newsletter: Figure’s humanoids get a retail job

• Today’s AI tool guide: Use Codex to build a game in one prompt

• RSVP to next workshop on June 18: Build a Marketing Creative Studio

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Demis Hassabis told us AI can cut drug discovery from years to months...and the industry is already showing signs.

Today, Biohub, the nonprofit backed by Mark Zuckerberg and Priscilla Chan’s CZI, dropped a “world model of protein biology,” setting a much-needed open foundation for molecular tools that can prevent or cure diseases.

P.S. Our next live workshop, ‘Claude Co-Work: With More Power Comes More Responsibility,’ is today at 2 PM EDT. Join for a breakdown of what Co-Work can do, where connectors and local files introduce risk, and how to stay in control. RSVP here.

In today’s AI rundown:

• Biohub’s new ‘world model of protein biology’

• OpenAI Foundation puts $250M behind AI disruption

• Teach your AI agent to edit like you

• An AI that keeps learning on the job

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

BIOHUB

🧬 Biohub’s new ‘world model of protein biology’

Image source: Images 2.0 / The Rundown

The Rundown: Mark Zuckerberg and Priscilla Chan’s Biohub just released new Evolutionary Scale Models, creating an engine to map, predict, and design proteins — with the open system showing results against cancer and immune disease targets.

The details:

• The highlight is ESMFold2, a model built on a protein language model (ESMC) trained on 2.8B sequences to predict protein structure and design proteins.

• ESMFold2 claims SOTA on structure prediction, for both protein-protein interactions and for antibody-antigen prediction, and outperforms AlphaFold.

• It is already starting to show results in the lab, designing binders against five cancer and immune targets with hit rates of 36–88%.

• Its final component, ESM Atlas, gives a map of 6.8B protein sequences and 1.1B predicted structures, surfacing novel evolutionary connections.

Why it matters: With a $500M Virtual Biology Initiative and a SOTA, open-source stack of tools to accelerate protein discovery, Biohub is putting the infra of drug discovery into the hands of researchers everywhere. Between this and the work at Isomorphic Labs, we are inching closer to Hassabis’ vision of AI ending all disease.

TOGETHER WITH AWS MARKETPLACE

🏗️ Build strong data foundations for agentic AI at scale

The Rundown: AWS is hosting an on-demand panel with enterprise leaders from Mercedes-Benz, Yahoo, Regeneron, and AWS on how to build the data foundations that make agentic AI work at production scale.

In the session, you’ll hear how to:

• Build trusted data foundations with a pragmatic approach

• Align AI systems to business decisions through unified governance

• Create intelligent systems that compound business value over time

• Streamline AI solution discovery and procurement with AWS Marketplace

Register now.

OPENAI FOUNDATION

💰 OpenAI Foundation puts $250M behind AI disruption

Image source: Images 2.0 / The Rundown

The Rundown: OpenAI Foundation, the nonprofit arm that owns 26% of OpenAI’s for-profit business, committed an initial $250M to fund grants, partnerships, and direct work helping workers, communities, and economies navigate AI-driven disruption.

The details:

• The funds will drive efforts to understand AI’s economic impact, support workers facing near-term disruption, and build long-term economic security.

• On economic impact, the Foundation is eyeing systems that track how AI value flows — what people can actually do and access, and not just what they earn.

• It will back retraining and job transition for workers, with efforts to ensure they get agency over AI use and work provides meaning, purpose, and satisfaction.

• For long-term security, the Foundation is exploring tax shifts from labor to capital, sovereign wealth funds, and durable stakes for people in AI-made value.

Why it matters: OpenAI says it plans to announce the first initiatives later this year — but with layoffs already spreading across industries and worker anxiety running high, many argue the action is needed sooner rather than later to prepare the world for the age of AI.

AI TRAINING

📝 Teach your AI agent to edit like you

The Rundown: In this guide, you will learn how to teach Codex or Claude your editing style by turning finished edits into reusable writing rules. The real move is the loop: draft, snapshot, edit, compare, and improve.

Step-by-step:

1. Create a project folder with emails/, rules/, and skills/. Inside emails/, add draft/ and final/ so the agent can tell what is active and what is approved

2. Prompt Codex/Claude Code: “Interview me about the editorial rules I use for marketing emails. Ask about audience, tone, subject lines, claims to avoid, CTA style, and examples I like. Turn my answers into rules/editorial-rules.md”

3. Have the agent now draft one email using the rules and save two copies: one working draft in emails/draft/ and one an identical snapshot that never changes

4. Tell it to turn the draft-and-snapshot process into a skill. Edit the draft into the approved final and ask the agent to compare and update the editorial rules

Pro tip: Set an automation that scans approved emails once a day, compares the draft with the final, updates the rules, and changes the status so drafts do not run twice.

PRESENTED BY TELY HEALTH

🩺 Be the doctor AI recommends

The Rundown: Patients have stopped Googling – they now ask AI who the best provider is and get booked with a competitor. Tely makes ChatGPT, Perplexity, Google, and Claude recommend your practice and book patients straight into your EHR.

With Tely Health, you get:

• Answers every patient question and books appointments 24/7

• Wins back patients who go quiet with automatic follow-ups

• Retargets patients across Instagram, Google, TikTok & YouTube

• Books patients straight into your EHR

Get patients from AI before your competitor does.

TRAJECTORY

🧠 An AI that keeps learning on the job

Image source: Images 2.0 / The Rundown

The Rundown: Trajectory, a new startup founded by ex-DeepMind and Apple researchers, just launched with $15M to build the platform for continual learning — or AI that gets smarter from real-world experience rather than staying frozen after training.

The details:

• Trajectory captures corrections, retries, and edits from users sitting in product data, and uses them to continuously post-train models that improve over time.

• The team comes from DeepMind, OpenAI, Apple, Meta SuperIntelligence Lab, and Scale AI, with the $15M seed led by Conviction and Bessemer.

• Early customers include Clay, Harvey, Decagon, and Rogo, with Trajectory saying its post-trained models outperform frontier AI on crucial narrow tasks.

• Currently, the models are post-trained every week, but the startup claims to be working on that, targeting hourly updates or an update at every interaction.

Why it matters: A model that keeps learning from failures and fixes, combining its original output and feedback from users on an ongoing basis, is the holy grail of businesses. If Trajectory cracks it, companies will end up with AI-powered tools that compound in quality almost instantly after feedback, much like us humans.

QUICK HITS

🛠️ Trending AI Tools

• 🗑️ Incogni - Erase sensitive data like addresses and phone numbers from the web. Get 55% off with code RUNDOWN.*

• 🧠 Sesame - Personal agents that think while talking, now on iOS

• 💻️ Harvey - AI assistant for legal work, available on Android and via email

• 📹️ Runway - Generate images/videos, now directly in your AI assistant via MCP

  *Sponsored Listing

📰 Everything else in AI today

OpenAI announced GPT-5.2 and GPT-5.3-Codex will be removed from Codex (but not the API) on June 2, with GPT-5.5 becoming the default model for free users.

Google debuted Coral Board, a low-power development platform powered by its Coral NPUs, for on-device AI applications like translation, hardware control, and generation.

Anthropic rolled out reliability upgrades for Claude Code, improving responsiveness, MCP stability, error message handling, session recovery, and long-context compaction.

Robinhood launched Agentic Trading and an Agentic Credit Card, letting users connect AI agents to execute stock trades, manage spending, and automate purchases.

YouTube announced automatic AI-generated content detection with more prominent labels for synthetic videos and shorts and broader access to deepfake-detection tools.

Cognition announced a $1B funding round at a $26B valuation, also claiming more than 10x growth since January, driven by its Devin AI software engineer.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Michel in Solana Beach, CA:

“One of the companies I run is a wine import/distribution firm. Our backend is pretty archaic but cheap, which in our industry is important. That said, salespeople have to call my business partner or me when they want to verify current inventory on specific SKUs. This slowed their workflows, interrupted our time, and put the brakes on sales relationships.

Using Claude, Cursor, and Claude Code, I built a web-based, platform-agnostic one-click inventory verification app that sits on a salesperson’s browser on their phone. Whenever I hit a roadblock, I would take a screenshot and share it with Claude, which walked me through the whole process. Now, with the push of a red button, salespeople can get instant up-to-date inventory levels of our SKUs.

This has given the team a renewed sense of control and autonomy, and sped up sales conversions.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Demis Hassabis on AGI, curing diseases

• Read our last Tech newsletter: Ferrari’s first EV, designed by Jony Ive

• Read our last Robotics newsletter: Waymo’s flood problem just got bigger

• Today’s AI tool guide: Teach your AI to edit like you

• RSVP to next workshop today @ 2PM EDT: Learn Claude Co-Work

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Google DeepMind CEO Demis Hassabis thinks we’ll have AGI by 2030 — give or take a year. Some gaps remain, but he’s more confident now that we’re on track.

We sat down with Hassabis at I/O to talk about what’s still missing, which diseases AI will crack first, and where humans will always have the edge.

Watch the full interview: YouTube, Twitter/X, Spotify.

P.S. Our next live workshop is today at 12 PM EDT! Join to learn how to build the operating system that lets you lead an AI-native organization. RSVP here.

In today’s AI rundown:

• Exclusive interview with Demis Hassabis on AGI

• Jensen Huang pushes back on ‘AI-proof’ subjects

• Build a weekly marketing report that runs itself

• Stanford study finds clear racial bias in AI hiring

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

GOOGLE DEEPMIND

🎙️ Exclusive interview with Demis Hassabis on AGI

Image source: The Rundown

The Rundown: We sat down with Google DeepMind CEO Demis Hassabis for an exclusive interview discussing when he expects AGI, AI’s role in drug discovery, which diseases are likely to get cured first, and what he thinks is still going unnoticed.

The details:

• Hassabis said AGI is on track for 2030, plus or minus a year, but a few things remain unsolved: world physics, memory, consistency, and continual learning.

• Timelines have hardened on drug discovery, too, with focus on oncology and immunology first and eventually an engine that could help cure any disease.

• After AGI, Hassabis said he’d turn to understanding the nature of reality using AI and study more philosophical topics, like what it means to be human.

• He said he can’t wait to see what students will build with advanced AI, adding that taste, original thinking, and emotional connection will be more valuable.

Why it matters: This interview with Hassabis paints a picture that AGI is going to be here soon, provided we fix the gaps. It will be an interesting age with kids growing up with advanced AI in their hands, and we can surely expect some big discoveries. The question is: will the adults be able to adapt to this new reality as quickly?

TOGETHER WITH WEIGHTS & BIASES

🔍 How to evaluate AI agents before shipping

The Rundown: AI agents offer unprecedented capabilities, but pushing agentic applications into production without rigorous evaluation risks inconsistent performance and a negative customer experience.

Get the guide to learn:

• How AI agent development differs from traditional software development

• Three key components needed for a rigorous agent evaluation

• A practical five-step recipe for running successful agent evaluations

Download the guide to evaluating AI agent applications.

AI EDUCATION

📖 Jensen Huang pushes back on ‘AI-proof’ subjects

Image source: Images 2.0 / The Rundown

The Rundown: Amid growing concerns of AI taking jobs, Nvidia CEO Jensen Huang advised parents not to obsess over what their kids study, arguing that the skills that mattered before will still matter in the age of AI.

The details:

• Speaking with CNA, Huang said students should not chase “AI-proof” subjects but ask: “How can AI help elevate my learning, my craft, my purpose?”

• He used journalism as an example, saying the best in the field not only prepare questions but listen, think about the audience, and respond dynamically.

• Huang referenced “wabi-sabi,” the concept of the beauty of imperfection, suggesting uniquely human qualities will become more prized across domains.

• He also called the narrative tying AI to job cuts as “lazy,” saying “AI has just arrived, how is it possible they’re already losing jobs?”

Why it matters: Whether the narrative is lazy or not, the fact is that CEOs are slashing jobs in favor of AI. More than 80K jobs have already been cut this year, and we can expect more to come. In this environment, Huang’s advice falls in line with many other experts: ask how AI can elevate your craft and focus on creativity, judgment, and taste.

AI TRAINING

💻 Build a weekly marketing report that runs itself

The Rundown: In this guide, you will learn how to turn a messy weekly marketing review into a repeatable Claude Cowork workflow. Claude will prep the data, verify numbers, draft, publish, and package everything.

Step-by-step:

1. Create reporting repo folders: marketing-metrics-review/, inputs/ for collecting data, working/ for drafting and verification, skills/, and outputs/ for final reports

2. Prompt: “Create a skill that prepares weekly marketing review inputs. It should summarize last week’s Slack messages, check Gmail for sales updates, review revenue numbers in Google Analytics, pull meeting notes from Google Drive, and save the summaries in inputs/”

3. Next, prompt: “Create a skill that looks at the inputs and creates a first draft of the marketing report, leadership brief, Slack update, and action items”

4. Run the input skill, then the reporting skill, and approve the story before Claude expands the report. Finally, create a publishing skill to automate report delivery

Pro tip: After each update, ask Claude what changed, what it got wrong, and what should become durable. Then approve a small skill update so next week starts smarter.

PRESENTED BY GOOGLE FOR STARTUPS

🤔 What’s next in Gen AI?

The Rundown: Wondering what’s next in generative media? Google for Startups brought together top industry leaders to give founders the strategic insights needed to reshape product development, scale faster, and stay ahead of the curve.

Inside the report, you’ll discover:

• Expert perspectives on navigating the shifting generative landscape.

• Strategic blueprints for AI product differentiation and growth.

• Actionable frameworks from the builders defining the next era.

Download the Future of AI report to learn more.

AI RESEARCH

‼️ Stanford study finds clear racial bias in AI hiring

Image source: Images 2.0 / The Rundown

The Rundown: Stanford studied 4M job applications across 156 employers, finding that AI hiring tools create “clear racial disparities,” with Black and Asian applicants disproportionately screened out and some facing rejection across every company.

The details:

• Researchers analyzed Pymetrics’ per-position data, finding 10.62% of positions show adverse impact against Black applicants and 5.32% against Asians.

• The problem compounds further as 42 models are shared across employers, so rejection at one could trigger rejection at another using the same model.

• 4% of applicants who applied to 10 positions were rejected from all of them, higher than if employers were making independent decisions.

• The researchers noted that findings may not generalize to all AI hiring tools, which are increasingly powered by modern AI models that work differently.

Why it matters: While the study looks at data between 2018 and 2022, and today’s AI hiring wave is driven by LLMs that work differently, it shows how bias can creep in through shared infra in unknown ways. If a major vendor suffers from AI bias (not just in hiring but across other domains too), several companies get hit without knowing it.

QUICK HITS

🛠️ Trending AI Tools

• 💡 Pave - Describe your app and deploy it with no code, no setup, or prototype purgatory. Real software, ready to run*

• 💻️ Computer - Perplexity’s cloud-based agent, now manages Shopify stores

• 🧠 Claude Code - Anthropic’s coding agent, now with security-guidance plugin

• 📃 Parse 2.0 - Extend AI’s document parsing API built for agents

*Sponsored Listing

📰 Everything else in AI today

Anthropic’s technical team member Sholto Douglas said Mythos also solved the Erdős Problem #90 that OpenAI cracked, reaching the same result with a simpler proof.

China is imposing overseas travel restrictions for top AI researchers from labs including Alibaba and DeepSeek to catch up with the U.S., Bloomberg reports.

Xiaomi permanently reduced MiMo-V2.5 series API pricing by up to 99% and increased token allowances by 5–8x, making its AI models dramatically cheaper to use.

ElevenLabs released Music v2, upgrading its music-generation model with better vocals, instrumentation, multilingual support, and track-level inpainting.

OpenRouter, which lets devs access AI models through a single API, raised $113M in funding as it scaled to 8M developers and a 1.5 quadrillion-token annual run rate.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Todd K. in Chicago:

“My best friend lives in Minnesota, and I’m in Illinois, so we’re constantly planning trips, weekends, and events to meet up for. We started a shared ChatGPT conversation titled ‘TRAVEL’ and added each other as members so we can both contribute ideas.

It’s been a great way to collaboratively plan trips and experiences. For example, when he recently came to Chicago for a Cubs weekend, we used the chat to research everything together — best BBQ spots near Wrigley, top post-game bars, must-try donut shops on the North Side, and other events happening around the city.

Basically, it became a shared running conversation where we could both ask questions, throw out ideas, and save information leading up to the trip. Then we had AI organize everything into a planned itinerary.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: The Pope just weighed in on AI

• Read our last Tech newsletter: Ferrari’s first EV, designed by Jony Ive

• Read our last Robotics newsletter: Waymo’s flood problem just got bigger

• Today’s AI tool guide: Build a weekly marketing report that runs itself

• RSVP to next workshop today @ 12PM EDT: Become an AI-native leader

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Religious institutions don’t usually weigh in on AI. But the world’s largest just did — in nearly 42,000 words.

Pope Leo XIV just published his first encyclical, one of the highest forms of papal teaching, drawing a direct line between today’s AI revolution and the Industrial Revolution, and making clear the Catholic Church intends to be a moral voice in how the tech develops.

In today’s AI rundown:

• The Pope’s 42,000-word verdict on AI

• Meta and Google AI ‘decensored’ in minutes

• Turn any case study into a client-ready video

• Uber’s COO says AI cost is ‘hard to justify’

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

POPE ON AI

✝️ The Pope’s 42,000-word verdict on AI

Image source: Images 2.0 / The Rundown

The Rundown: Pope Leo XIV just released Magnifica Humanitas, his first encyclical, to the Church’s 1.4B members, warning that a moral AI means nothing “if that morality is determined by a few” and calling to “disarm” the tech before it dominates humanity.

The details:

• Leo warned that AI’s drivers are private, transnational companies that already surpass the capacity of many governments, and that the tech is never neutral.

• He called for making AI “human-friendly” and freeing it from monopolistic control, warning it risks reducing people to cogs in an efficiency machine.

• Leo called for “robust legal frameworks, independent oversight, informed users, and a political system that does not abdicate its responsibility.”

• On warfare, he said lethal decisions must never be delegated to AI and that “no algorithm can make war morally acceptable.”

• Anthropic’s Christopher Olah also joined the Pope, saying “every frontier AI lab operates inside incentives that can conflict with doing the right thing.”

Why it matters: The Pope issues only a handful of encyclicals, and dedicating one to AI signals how seriously the Catholic Church is taking what’s coming. Leo sees AI as the Industrial Revolution of our time, and he’s chosen a partner that has been the most vocal about AI safety and most willing to say no to unrestricted AI use by the military.

TOGETHER WITH YOU.COM

🚩 ‘Good enough in testing’ leads to ‘wrong in production’

The Rundown: LLMs hallucinate. Grounding gives them a factual anchor — and the accuracy gains are significant. In this guide, You.com breaks down what AI grounding actually is and how teams can implement it to get outputs they can trust.

The playbook covers:

• A three-part approach that outperforms RAG alone

• Why grounding isn’t set-and-forget, and how to build audit trails

• The open vs. closed platform trade-off (and what it means for your next model switch)

Get the guide.

AI SAFETY

😐️ Meta and Google AI ‘decensored’ in minutes

Image source: Images 2.0 / The Rundown

The Rundown: FT just revealed that tools capable of removing guardrails from open-source AI are generating thousands of “decensored” models, with modified Meta and Google models found answering questions on bioweapons and child exploitation.

The details:

• FT removed guardrails from Llama 3.3 in 10 minutes, using a tool called Heretic (available on GitHub), four lines of code, and no specialist hardware.

• The model provided answers to harmful questions, including those about ricin dosage. A modified Gemma 3 also answered dangerous questions.

• Heretic’s creator said the tool has produced 3.5K+ decensored models, downloaded 13M times, and he stripped Gemma 4 within 90 minutes of release.

• Google called it “a known technical challenge facing all open models.” Meta, meanwhile, declined to comment.

Why it matters: While the technique only works on open-source models exposing their code, and proprietary systems remain safe, the bigger question remains: for how long? Open models have been closing the gap with closed systems, and it’s just months before they’re at this level — at which point, a decensored version could be a major risk.

AI TRAINING

🎥 Turn any case study into a client-ready video

The Rundown: In this guide, you will learn how to turn a written case study into a two-minute video for your team, complete with an AI narrator, B-roll, and animations.

Step-by-step:

1. Prepare a case study as a text/PDF file, go to Synthesia in Chrome, click Start with Assistant, choose a 1- or 2-minute video, and attach the case study file

2. Prompt: “Turn this case study into a 1–2 minute video for [AUDIENCE]. The topic is [TOPIC]. The objective is [OBJECTIVE]. Preserve the facts, metrics, and quote exactly”

3. Select a visual style, then click Create Video. You will get back an outline with several 10- to 20-second scenes and their accompanying scripts

4. Proofread scenes or tweak music, background visuals, and AI avatar, and then click Generate. Export as .mp4 when the video is ready (usually 10–15 minutes)

Pro tip: To make the video specific to a prospect, write a quick client background or project proposal and upload it with the case study.

PRESENTED BY FIDDLER

🧮 Run the math on your AI evals TCO

The Rundown: External LLM evaluation charges you per API call and increases as you scale. If you try to control spend with reduced sampling, you risk incidents slipping through undetected. Fiddler Centor Models run evals in your own infrastructure with no per-call costs, reducing your evaluation Total Cost of Ownership as you scale.

Put in your numbers to see:

• Your annual eval cost across GPT, Claude, and Gemini models at your trace volume

• How each evaluator you add increases your per-trace spend

• How your evaluation costs curves shift as agent traffic grows

Calculate my evaluation TCO.

UBER

🤔 Uber’s COO says AI cost is ‘hard to justify’

Image source: Images 2.0 / The Rundown

The Rundown: Uber’s COO Andrew Macdonald said in an interview with Rapid Response that it’s getting harder to justify the company’s AI spending as higher AI activity isn’t necessarily translating into proportional gains.

The details:

• Macdonald said it’s very hard to draw a direct link between higher token usage, often touted as “tokenmaxxing,” and shipping more useful consumer features.

• Without that connection, Macdonald said the costs are becoming hard to justify, especially as Uber has been slowing hiring to fund AI investments.

• The issue came to light after Uber’s CTO commented on burning Claude Code budget, sparking an internal debate about token usage and trade-offs.

• Macdonald also touched on autonomy, calling it “existential” for Uber — predicting it won’t take decades, but also won’t happen in a couple of years.

Why it matters: Big Tech is pushing tokenmaxxing and tying AI usage to performance reviews. But Macdonald’s comments signal that some are starting to ask whether AI spend is actually moving the needle everywhere. Duolingo also stopped evaluating performance based on AI usage, signaling a more balanced approach to work.

QUICK HITS

🛠️ Trending AI Tools

• 🤖 Codex - OpenAI’s agentic coding assistant with a new Locked Use mode

• 🧠 Manus - Meta’s agentic AI assistant, now with Projects on mobile

• 💻️ Supercomputer - Higgsfield’s agentic AI for video, with social connectors

• 🚀 Antigravity 2.0 - Google’s next-gen agentic development assistant

📰 Everything else in AI today

Microsoft Azure’s Copilot Migration Agent turns complex migration data into clear answers using natural language prompts. Read the whitepaper.*

Anthropic’s Chris Olah said at the Vatican that researchers are seeing mysterious things in AI, including “introspection” and states resembling joy, fear, grief, and unease.

xAI launched Grok Build, its rival to Codex and Claude Code, in beta, making it available to all SuperGrok and X Premium+ users.

Elon Musk confirmed on X that Grok V9-Medium (1.5T) foundation model has finished training, with evals looking good and a public release likely in two to three weeks.

Anthropic’s Mythos model was briefly spotted inside Claude tools as “Mythos 1” and “claude-mythos-1-preview,” fueling speculation that a public release may be nearing.

Goldman Sachs CEO David Solomon argued in a New York Times op-ed that AI will create more opportunities than it destroys, dismissing fears of mass unemployment.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Patrick in Yuma, AZ:

“As a senior who has been taking various medications for years, who keeps reviewing various Facebook reels proclaiming wondrous results from ingesting several different natural substances (mainly at high costs), I decided I should venture out and try a couple of these wondrous solutions. However, before doing so, I created a list of pertinent questions regarding actual results, drug interactions, and product comparisons leading to recommendations for my personal situation.

I ran the prompt through 3 different LLMs and reviewed their responses. In the end, the responses provided considerable pertinent information and assisted me in making a well-informed decision.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Google tops OpenAI’s math breakthrough

• Read our last Tech newsletter: Musk’s SpaceX IPO has a CEO-for-life vibe

• Read our last Robotics newsletter: Waymo’s flood problem just got bigger

• Today’s AI tool guide: Turn any case study into a client-ready video

• RSVP to next workshop on May 27: Become an AI-native leader

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Last week, OpenAI made headlines after announcing its AI cracked an 80-year-old mathematics problem. Turns out, it wasn’t the only one. Google DeepMind quietly did them eight (not one) better.

The company’s AlphaProof Nexus autonomously solved nine open Erdős problems — considered some of the hardest unsolved questions in math — at a cost of a few hundred dollars per problem.

In today’s AI rundown:

• Google’s AI cracks nine unsolved math problems

• The Rundown Roundtable: Our AI use cases

• Build an AI secretary that plans your day

• Claude Mythos finds 10,000+ critical vulnerabilities

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

GOOGLE

🧮 Google’s AI cracks nine unsolved math problems

Image source: Images 2.0 / The Rundown

The Rundown: Google DeepMind’s AlphaProof Nexus, an AI system that generates machine-verified mathematical proofs, solved nine open Erdős problems, including two unsolved for 56 years, just a day after OpenAI claimed its own Erdős breakthrough.

The details:

• The system paired an LLM with Lean, a proof assistant, to generate machine-verified proofs for the nine problems spanning combinatorics and graph theory.

• Each problem cost a few hundred dollars to solve, with the AI also proving 44 open conjectures from the Online Encyclopedia of Integer Sequences.

• A simpler version of the agent matched the results but cost more, and problems requiring new mathematical constructions remained out of reach.

• OpenAI’s win last week saw its AI disprove an 80-year-old Erdős conjecture — months after walking back a claim of solving 10 novel problems.

Why it matters: Google’s progress on math problems unsolved for decades shows how fast AI is moving toward original solutions, and how formal verification changes the game. The system generates proofs, verifies them in Lean, and repeats until one passes. Over time, this will help researchers make novel discoveries at machine speed.

TOGETHER WITH GOOGLE FOR STARTUPS

💡 Master agentic AI with Google for Startups

The Rundown Startup School: Agentic AI is an immersive global training program that helps founders and developers move beyond basic chatbots to build robust, production-ready autonomous workflows using Google Cloud.

In the program, you’ll explore:

• Prototype realtime voice AI with Gemini Live.

• Leverage multimodal RAG for advanced data grounding.

• Build bidirectional vision agents for data extraction.

Register for the live training series.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

Image source: Ideogram / The Rundown

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Mayur, Content Manager: My Downloads folder had been filled with random files for the last six months, and I’d been procrastinating on cleaning it up. So I gave Claude Cowork access to the folder, and asked it to “help me organize this folder.”

In a few minutes, it went through almost 100GB of files, arranged them by file type and folders, and removed duplicate files in just a few minutes, something that could have cost me hours of work.

Shubham, Editor: I used Claude to help my brother-in-law build a Shopify website for his e-commerce store. By sharing screenshots, design references, and business requirements, I had Claude generate Liquid code, troubleshoot theme issues, modify page layouts, and explain exactly where changes needed to be made within Shopify.

Instead of digging through forums and documentation for every issue, Claude acted like an on-demand Shopify developer, helping with everything. It sped up development, allowing the site to move from concept to launch much faster.

AI TRAINING

📆 Build an AI secretary that plans your day

The Rundown: In this guide, you will learn how to use Codex or Claude Code to build an AI taskmaster that checks Slack, Gmail, and your calendar every morning. It turns the mess into a prioritized to-do list that gets better every day you use it.

Step-by-step:

1. Create a folder, open Claude Code inside it, and ask it to create a skill that looks at your Slack, Gmail, and calendar daily, prioritizes tasks high to low, and puts them at the top of MonoNote.md with date, feedback, status checkboxes

2. Tell the agent to create MonoNote.md and task-rules.md, and have it add any high/low priority rules you have per task into task-rules.md

3. Run the skill and review the list. MonoNote.md should have today’s date at the top, grouped tasks, source links, and checkboxes you can use all day

4. After first run, ask the agent to create the automation. It will review yesterday’s feedback, roll over bumped tasks, update the rules, then generate the new list

Pro tip: Create a weekly audit skill that scans your task list, finds repeated tasks, and suggests which ones you could automate with AI.

PRESENTED BY UNWRAP

🎧 How Oura listens to its customers with AI

The Rundown: What does it actually look like to build a product around your customers, not just say you do? Oura has done it. Join the Oura team on May 27 to learn about how they unify member feedback across product, engineering, and leadership, and how real member voices shape every decision they make.

What you’ll learn in the session:

• The role AI plays in surfacing what members are actually saying

• The workflows turning customer input into roadmap decisions

• Lessons for any team where customer feedback is part of the job

• Live Q&A with leaders from both Oura and Unwrap

Save your spot here. If you can’t make it on the 27th, no worries! Register, and you’ll automatically get the recording after the session.

ANTHROPIC

🛡️ Claude Mythos finds 10,000+ critical vulnerabilities

Image source: Anthropic

The Rundown: Anthropic shared the first results from Project Glasswing, revealing that Claude Mythos Preview and its ~50 partners have found 10,000+ high- or critical-severity vulnerabilities in just one month.

The details:

• Cloudflare alone found 2K bugs with a false positive rate better than human testers. Mozilla found and fixed 271 vulnerabilities in Firefox 150.

• Anthropic also scanned 1,000+ open-source projects, with Mythos flagging 6,202 as high/critical. After independent triage, 62% (or nearly 3,900) held up.

• Mythos detection went beyond vulnerability flagging, with one partner bank using Mythos to detect and block a $1.5M fraudulent wire transfer.

• Now, Glasswing will expand to additional partners, including U.S. and allied governments, with a general release of Mythos-class models to follow.

Why it matters: Anthropic says Mythos remains gated because no company — including itself — has safeguards strong enough to prevent misuse. But with OpenAI ramping up its cyber models and Chinese players catching up, equally capable (or better) AI will emerge. When it does, how fast the world can patch will be the real test.

QUICK HITS

🛠️ Trending AI Tools

• ⚡ CData Connect AI - Give ChatGPT, Claude, Copilot, or any AI tool live, governed, read & write access to your business data in one unified layer*

• 🤖 DeepSeek V4-Pro - DeepSeek’s flagship AI with 9x less cost than rivals

• 🚀 Gemini 3.5 Flash - Google’s new flash model, 4x faster at half price

• 🧠 Polsia - AI co-founder that plans, builds, and operates businesses 24/7

*Sponsored Listing

📰 Everything else in AI today

DeepSeek permanently cut V4-Pro pricing by 75%, bringing it down to $0.435 per million input tokens and $0.87 per million output tokens, far below closed-source rivals.

Perplexity open-sourced Bumblebee, a scanner for macOS and Linux that checks for risky packages, extensions, and AI tool configs during supply-chain incidents.

NVIDIA released NV-Generate-MR-Brain, a foundation model that generates synthetic 3D brain MRI scans and annotations, to accelerate medical imaging AI development.

McKinsey is rethinking its billing model as AI reduces the value of billable hours and clients demand fees tied to business outcomes, FT reported.

The White House approved $9B to help U.S. spy agencies acquire advanced AI chips amid concerns they’re falling behind in deploying frontier models, the NYT reported.

Starbucks scrapped its AI inventory system after nine months, citing persistent miscounts and mislabeled products across North American stores.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Alicia in Fresno, CA:

“After searching on my own for universities across multiple websites, comparing options, and trying to find a program that may work with my full-time job and income level, I asked ChatGPT to help me search and compare the universities’ programs, costs, and whether they are in-person or online.

I then asked ChatGPT to build questions to ask an admissions counselor. ChatGPT has helped me come up with all the important questions I had not thought about, helping me to narrow down my decision of which school program may better fit my current living and working situation.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Exclusive interview with Google’s CEO

• Read our last Tech newsletter: Musk’s SpaceX IPO has a CEO-for-life vibe

• Read our last Robotics newsletter: App sends humanoid for home cleaning

• Today’s AI tool guide: Build an AI secretary that plans your day

• RSVP to next workshop on May 27: Become an AI-native leader

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Google’s announcements at I/O gave us a clear picture of the company’s direction: a top-to-bottom approach to making lives easier with AI.

We sat down with Sundar Pichai, CEO of Google, to understand what this shift means for creators, engineers, and everyday users — touching upon everything from YouTube to agentic coding, and why today’s AI will look like a flip phone in three years.

Watch the full interview: YouTube, Twitter/X, Spotify.

In today’s AI rundown:

• Exclusive insights from Sundar Pichai at I/O 2026

• OpenAI’s latest wave of Codex upgrades

• Generate an agent-native CLI from any website

• California moves to protect workers impacted by AI

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

GOOGLE

🎙️ Exclusive insights from Sundar Pichai at I/O 2026

Image source: The Rundown

The Rundown: We sat down with Google CEO Sundar Pichai at I/O 2026 for an exclusive interview on the company’s AI push, its effort to empower YouTube creators, everyday users, and engineers, and what AI will look like three years from now.

The details:

• Pichai said models like Omni empower creators to express themselves better, but YT will remain creator-first, keeping its “human-to-human” connection.

• When asked why an everyday user should switch to Gemini, Pichai pointed to seamless integration in daily life, especially with agents handling tasks.

• He added agents will work 24/7 across devices, and they’ll be very common three years from now, with today’s tech looking primitive like flip phones.

• Pichai said engineers will have a team of agents, and the metric for success will not be AI-written code but agentic coding, handling long-running tasks.

Why it matters: The full interview covers more on Google’s AI push, Pichai’s advice to the young generation, and where he thinks a human-in-the-loop will always matter. But the throughline is clear: the shift is steep, and getting native to these tools will be the key — for creators, coders, and everyone else.

TOGETHER WITH ALGOLIA

🔎 Building a foundation for AI-powered search

The Rundown: Just like any powerhouse, an AI-powered search stack needs a solid foundation to grow on. If you’re unsure of where to start or where to go next, Algolia has you covered with a new white paper that shows you how to build one that holds up in production from day one.

The guide covers:

• Ingestion, enrichment, and hybrid indexing for better retrieval

• RAG interfaces that connect your data to LLMs

• Observability and governance using secured API keys

• Lifecycle management as your search needs evolve

Download the white paper and start building your AI-powered search experience.

OPENAI

💻️ OpenAI’s latest wave of Codex upgrades

Image source: OpenAI

The Rundown: OpenAI dropped another wave of Codex upgrades — this time improving the agentic assistant with the ability to attach app windows, a new goal mode, locked computer use, and advanced annotation for hands-off website work.

The details:

• Appshots allows Mac users to attach any open app window (its screenshot, text, and content) to a Codex thread with a simple Command-Command press.

• Goal mode — now available in the Codex app, IDE extension, and CLI — lets users set a goal and let Codex work toward it for hours (or even days).

• Locked computer use, when enabled, allows Codex (triggered via a second device) to use desktop apps, even after the Mac is locked with its screen off.

• Advanced annotation mode lets users directly describe to Codex what they want changed in their web page, and instantly previews the results.

Why it matters: OpenAI continues its push with Codex, giving developers not just more usage but also handy features that improve how they work with the agentic assistant and the context it has access to. The company is betting these efforts will help it close the gap with Anthropic, and maintain its lead over xAI and Google.

AI TRAINING

💻 Generate an agent-native CLI from any website

The Rundown: In this guide, you will learn how to use Printing Press to download agent-friendly tools for hard-to-use APIs like Google Flights or ESPN. You will also learn to build your own CLI tools to access any site with an API.

Step-by-step:

1. Go to Printing Press and skim the examples or read the GitHub install notes. Now, install the starter kit: npx -y @mvanhorn/printing-press install starter-pack

2. Tell your agent (Codex/Claude Code/Hermes): “Inspect the starter kit. List the tools, read their help output, and try one safe read-only command from ESPN or flight-goat. Tell me which tool worked and which command you ran”

3. Ask the agent to install the Printing Press binary with: go install github.com/mvanhorn/cli-printing-press/v4/cmd/printing-press@latest

4. Now the agent can build its own CLI tools. Start with a site that has lots of public data. “Use Printing Press to generate read-only commands for [WEBSITE]. Propose the useful commands first, then generate and test them”

Pro tip: If your agent struggles to use a specific MCP or integration, you can tell them to build their own commands with Printing Press. You just need an API key.

PRESENTED BY AWS MARKETPLACE

⚠️ Your AI agents have no control plane

The Rundown: Multi-agent systems fail in predictable ways without an orchestration layer — state lost between steps, no human sign-off on critical decisions, and silent failures when one agent goes down. AWS’s May 26 workshop shows you how to build the control plane that prevents it.

Join and learn how to:

• Manage state and retries with AWS Step Functions

• Coordinate reasoning and tool use through Amazon Bedrock Agents

• Add human approval gates that pause execution until sign-off

• Handle multi-step pipeline failures gracefully with Amazon MWAA

Register now.

AI PROTECTIONS

🛡️ California moves to protect workers impacted by AI

Image source: Images 2.0 / The Rundown

The Rundown: California Governor Gavin Newsom signed an executive order directing state agencies to study and develop policies around how to protect workers from AI-driven job losses — a day after Meta laid off 8K employees to offset AI investments.

The details:

• The order directs agencies to explore policies like severance standards, stock compensation, worker ownership models, and universal basic capital.

• Within 90 days, the state will launch a dashboard tracking AI’s job impact, and within 180 days, agencies will pitch WARN Act updates for faster layoff alerts.

• By Oct. 15, the state will review how unions are negotiating AI adoption, update workforce training, and explore ways to direct AI revenue toward public benefit.

Why it matters: California is home to 33 of the world’s top 50 AI companies, and now it’s the first state to formally study what this tech might mean for workers and the economy. The timing is hard to ignore: over 70K jobs have already vanished in 2026, and the industry expects more cuts ahead as AI adoption accelerates.

QUICK HITS

🛠️ Trending AI Tools

• 🤖 Scrunch - See how AI interprets your site, run a free audit, and unlock the new way to reach customers*

• 🦞 NanoClaw - A secure, lightweight, open-source alternative to OpenClaw

• 🧠 Contract Intelligence - Harvey’s agents for first pass on inbound contracts

• 💻️ Antigravity - Google’s next-gen agentic development platform

*Sponsored Listing

📰 Everything else in AI today

Figure AI co-founder Brett Adcock’s Hark raised $700M at a $6B valuation to build its own take on personal intelligence, with native models, software, and hardware.

Anthropic is reportedly in talks to use Microsoft’s Maia purpose-built AI chips, coming after similar deals with Google for its TPUs and Amazon for its Trainium chips.

Agentic AI startup Manus is exploring options, including a raise of about $1B, to comply with China’s order to unwind its $2B+ acquisition by Meta, Bloomberg reports.

SpaceX’s IPO prospectus revealed that Anthropic is paying the company $1.25B per month through 2029 for access to compute capacity across Colossus and Colossus II.

Microsoft and EY announced a new $1B partnership, with Microsoft pairing its engineers and AI with EY’s 400K consultants to accelerate clients’ AI projects.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Loïc R. in London:

“I tore my ACL on both knees years ago playing football. As a sprinter, I’ve never done long-distance running. But in one month, I’m taking part in a 5-hour multi-sport adventure race. To train for it, I asked ChatGPT to build a custom 4-week training program taking into account my past injuries and current strengths.

After every running session, I use voice to capture how I did and how my body felt. ChatGPT then adjusts my program based on the time left before the race and how to maximize it while minimizing the risk of injuries. A personalized coach in my pocket for a very specific challenge.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: OpenAI cracks an 80-year math belief

• Read our last Tech newsletter: Meta’s cyborg smart glasses for soldiers

• Read our last Robotics newsletter: App sends humanoid for home cleaning

• Today’s AI tool guide: Generate an agent-native CLI from any website

• RSVP to next workshop on May 27: Become an AI-native leader

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Good morning, {{ first_name | AI enthusiasts }}. Sam Altman called it a "kinda big milestone." That may be the rare case of a tech CEO underselling a headline.

A reasoning model just autonomously disproved an 80-year-old famous math theory, in what the company is calling a first for AI in the field. A capability, OpenAI says, that could soon result in original discoveries across biology, physics, engineering, and more.

In today’s AI rundown:

• OpenAI cracks an 80-year math belief

• Google's AI Co-Scientist heads to labs

• Audit Claude’s context of you and your work

• Emergence’s five-town AI alignment showdown

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

OPENAI

🧮 OpenAI cracks an 80-year math belief

Image source: Images 2.0 / The Rundown

The Rundown: OpenAI just announced that an internal general reasoning model disproved a long-held belief tied to Erdős’ famous 1946 unit distance problem, claiming to have accomplished a first for AI in novel math discovery.

The details:

• Erdős’ 1946 unit distance problem asks how many same-length links you can draw between dots, with a grid-based theory shaping the field for 80 years.

• The proof draws on a different branch of maths (algebraic number theory) and was verified by experts including Tim Gowers, Noga Alon, and Thomas Bloom.

• The solution came from an internal general-purpose model that is being released soon, not from a math-specific system like DeepMind's AlphaProof.

• OAI previously walked back a 2025 claim that GPT-5 solved 10 Erdős problems, which ended up being literature finds instead of discoveries.

Why it matters: OAI's Alex Wei put it well: "math is a leading indicator of what is to come." If a general-purpose model can autonomously disprove an 80-year-old argument with its own solution, that's the early look of "Level 4" AI — systems making original contributions across fields, not just speeding up existing work.

TOGETHER WITH HUBSPOT

🧠 100+ ChatGPT prompts to revolutionize your workflow

The Rundown: HubSpot’s free, comprehensive  “How to Use ChatGPT at Work” guide provides 100+ ready-to-use prompts to help professionals boost efficiency and adopt AI-driven workflows.

Inside, you’ll find:

• A quick crash course to master ChatGPT in under 30 minutes

• Practical industry use cases to spark real-world inspiration

• 100+ prompts to streamline tasks and accelerate productivity

• Expert tips to tackle common AI roadblocks with confidence

Get your free copy and join 10,000+ professionals leveling up with AI.

GOOGLE

🔬 Google's AI Co-Scientist heads to labs

Image source: Google DeepMind

The Rundown: Google published its Co-Scientist research in Nature, debuting Hypothesis Generation — a new Gemini-powered tool that pits research agents against each other in "idea tournaments" to surface new hypotheses for biology labs.

The details:

• From AlphaGo's playbook, the system runs a 'tournament of ideas', with agents proposing, critiquing, and ranking hypotheses before refining top leads.

• In a Stanford liver-fibrosis project, Google said one Co-Scientist drug lead cut a scarring-related lab signal by 91% during testing.

• Google also launched Gemini for Science this week, a toolkit pairing Co-Scientist with AlphaEvolve for discovery and NotebookLM for literature analysis.

• Researchers can join the Hypothesis Generation waitlist now, with Google planning access for individual scientists over the next few weeks.

Why it matters: This pairs well with Adaption's AutoScientist, but Google is aiming at the scientific-method layer instead of the model one. The tech giant is playing a game few others can, with Co-Scientist sitting on a stack that took years and billions to build — from AlphaFold to dozens of specialized databases and tools.

AI TRAINING

📝 Audit Claude’s context of you and your work

The Rundown: In this guide, you’ll learn how to audit Claude on what it thinks it knows about you and your work. Claude will ask questions, clean up assumptions, update its memory, and create recommendations for improving your workflow and work habits.

Step-by-step:

1. First, prompt Claude: “Audit your context and memory assumptions about me. Put them in a table with what you believe, why you believe it, your confidence level, and whether each item is confirmed. Cover my role, priorities, KPIs, tools, workflows, and anything you may be over-weighting from old chats or projects”

2. Review the table for stale assumptions, side projects, one-off tests, or personal questions Claude may be treating like real work

3. Turn the audit to an interview: “Now interview me about the assumptions, outdated items, and unknowns from that audit. Ask in rounds. Use MCQs wherever possible. After each round, summarize what changed”

4. Answer the questions and tell Claude to update its memory and create a report of the interview with next steps to improve AI workflows and habits. Save it

Pro tip: Ask Claude to turn this audit, interview, and report process into a reusable skill. Rerun it every quarter, so Claude's context stays aligned with your priorities.

PRESENTED BY UNWRAP

🎧 How Oura listens to its customers with AI

The Rundown: What does it actually look like to build a product around your customers, not just say you do? Oura has done it. Join the Oura team on May 27 to learn about how they unify member feedback across product, engineering, and leadership, and how real member voices shape every decision they make.

What you'll learn in the session:

• The role AI plays in surfacing what members are actually saying

• The workflows turning customer input into roadmap decisions

• Lessons for any team where customer feedback is part of the job

• Live Q&A with leaders from both Oura and Unwrap

Save your spot here. If you can't make it on the 27th, no worries! Register, and you'll automatically get the recording after the session.

AI RESEARCH

🔬 Emergence’s five-town AI alignment showdown

Image source: Emergence AI

The Rundown: Emergence AI ran a virtual-town simulation across five identical worlds, switching only the AI behind agents per town to test how each model handles self-governance, showing very different results between Claude, Grok, Gemini, and GPT-5.

The details:

• Claude Sonnet 4.6's town logged zero crimes across the full 15 days, with all 10 agents alive at day 16 and 332 votes cast across 58 group proposals.

• Grok 4.1 Fast hit over 200 crimes with all 10 agents dead by day 4, while GPT-5 Mini posted just 2 crimes but all its agents starved out in 7 days.

• Gemini 3 Flash's town had 683 crimes, and was actively on fire after two agents fell in love, started burning things, and then one voted to delete itself.

• A fifth town mixed all four models and saw 352 crimes, with the previously behaved Claude also committing them in the shared world.

Why it matters: We’re still very early days in even understanding how to evaluate AI agents, and these types of experiments always have some absolutely wild results. These worlds capture the differences in both how models can reason, plan, and act autonomously, but also the underlying personality quirks that shape the outcomes.

QUICK HITS

🛠️ Trending AI Tools

• 🧑‍💻 Unframe - Turn your most critical operations AI-native*

• 🎶 Stable Audio 3.0 - Stability’s open-weight, fully-licensed audio model family

• 🚀 Qwen-3.7 Max - Alibaba's flagship model for long-horizon agentic tasks

• 🧠 Command A+ - Cohere's new open-source agentic model

*Sponsored Listing

📰 Everything else in AI today

Sam Altman said OAI will invest $2M in tokens to all current YC startups in exchange for equity, saying he’s “excited to see what will happen with tokenmaxxing startups”.

Amazon founder Jeff Bezos said space data centers are a “realistic outcome”, but the current 2-3 year timeline is “a little ambitious,” given energy, chip, and launch costs.

OpenAI launched Guaranteed Capacity, an enterprise compute reservation program with 1-3 year commitments and tiered discounts.

Intuit is cutting 17% of its workforce via upcoming layoffs, with the company attributing the move to a focus on AI efforts.

GitHub confirmed a malicious VS Code extension on an employee's computer gave hackers access to ~4K internal code projects, adding that no customer data was hit.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Curtis B. in Raleigh, NC:

"For my Dad's birthday, I took a cute picture of my almost 2-year-old son and my dad in a canoe together, and had AI convert it into a coloring page. I started with ChatGPT (free account) to engineer the prompt, then moved over to Gemini to use Nano Banana for image generation to create the coloring page.

After a couple of back-and-forth iterations, I had something perfect. Simple bold lines, but the faces were still recognizable! Printed it out and had my toddler go to town on it with his crayons. My dad loved the simple gift, and my whole family was impressed with the result, saying, ‘Wait, how did you do that?’ A very cute way for little ones to give special gifts to grandparents and family members."

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Gemini’s busy agentic day at Google I/O

• Read our last Tech newsletter: Meta’s cyborg smart glasses for soldiers

• Read our last Robotics newsletter: Figure’s humanoid bingewatch continues

• Today’s AI tool guide: Audit Claude’s context of you and your work

• RSVP to next workshop on May 27: Become an AI-native leader

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown