Unicorner

Flexion

Mon, 18 May 2026 15:00:00 +0000

Missed the news?

Last week, we announced our Boston + Tech Week lineup, including AI Day #NYTechWeek - a day + night experience at NY Tech Week built around the people and products shaping the future.

We’re hosting three firesides with founders and executives from Bilt, Bolt.new, and Graphite (acquired by Cursor), followed by a 30-person dinner for product, design, and engineering leaders, and capped with an afterparty featuring Ian Asher.

See you there.

Tues. May 26 - Founding GPs VC Omakase (Boston)
Wed. May 27 - AI Champions Dinner (Boston)
Tues. June 2 - Founding GPs VC Omakase (New York)
Wed. June 3 - AI Day (New York)
Tues. June 23 - AI Champions Dinner (San Francisco)

Ethan and Arek 🦄

Flexion is building the autonomy stack that gives humanoid robots the ability to understand tasks in natural language, plan and execute multi-step workflows, and adapt to real-world conditions with no human in the loop. The platform is designed to run across hardware from different humanoid manufacturers without modification.

Check it out: flexion.ai

Flexion offers annual per-robot software licensing. Flexion sells to OEM robot manufacturers and enterprise operators, positioning itself as a horizontal intelligence layer across the humanoid ecosystem rather than a vertically integrated hardware company. Think of Microsoft's strategy as opposed to Apple’s.

Raised a $50 million Series A led by DST Global Partners, with participation from NVentures (Nvidia's venture arm), Redalpine, Prosus Ventures, and Moonfire Ventures
Actively piloting with major OEM customers
45 employees across the Zurich R&D headquarters, with ongoing expansion efforts in the Bay Area

Flexion is actively hiring across AI engineering and hardware engineering in Zurich and the Bay.

The Flexion team came from the same ETH Zurich lab, each holding different academic titles. Simultaneously, Rudin and Hoeller were working at Nvidia with simulation and reinforcement learning tools for robotics.

While working directly with companies building humanoids, they kept seeing the same components built and obstacles hit independently. They realized the compute and data flywheel that unlocked LLMs had no equivalent in robotics. Flexion was founded to give humanoid manufacturers and operators a ready-made intelligence layer, so that they could focus on their hardware and deployment rather than reinventing the autonomy stack from scratch.

Bringing AI models into the physical world is the next AI frontier.

Of the infinitely possible robot form factors, building generalized humanoid robots is the most ambitious. Fortunately, on the hardware front, the technology has quickly improved, with actuators that are more precise, battery systems that last longer, mechanical design that has become more robust, and bipedal platforms that can now navigate real environments and perform structured tasks at a level that would have seemed far-fetched five years ago. Instead, the bottleneck holding back deployment at scale is intelligence.

Robotic operating systems have been historically brittle. Humanoid robots look impressive in demos, but few can perform outside these controlled settings. With task-specific code and teleoperation pipelines, prior systems had hand-engineered behaviors that required significant reprogramming when conditions shifted. At scale, that created a landscape where every manufacturer was independently building an autonomy stack from scratch, often similar to each other, but not general enough to ensure high robot uptime.

Flexion is building this layer. Its autonomy stack handles natural language task understanding with multi-step workflow execution, creating real-world adaptation across humanoid robots regardless of their physical configuration. That means robots with differing numbers of joints or sensor layouts across manufacturers can work with Flexion. The company has correctly identified and bet that a horizontal software platform sitting above the hardware competition captures value regardless of which robot body wins in any given segment.

Flexion is building the software layer that enables humanoid robots to perform real-world tasks with minimal human input.

Zooming out, humanoid robotics has crossed a threshold. Roughly 50 companies have declared ambitions in humanoid robotics, but fewer than ten have advanced to scaled pilots or precommercial deployments. Meanwhile, robotics-related venture investment exceeded $10.7 billion globally as of late 2025, and the global humanoid robotics market could hit $38 billion by 2035.

Looking even further, others forecast the total market could surpass $5 trillion by 2050. Flexion's per-robot licensing model scales directly with the number of robots in the field, a number expected to reach nearly 3 billion by 2060, and the inbound interest from OEMs, enterprise operators, and the aforementioned investors, including Nvidia’s venture arm, is signal that the market has already recognized the gap being filled.

The conversation has shifted from whether humanoids will be useful to when and at what scale. The defining challenge ahead is building the data flywheel fast enough. More robots deployed means more training data, which leads to stronger models, which in turn attract more deployments. Whoever builds the shared intelligence layer first captures the compounding advantage, and Flexion has the architecture and the people to make a serious run at it.

The evolution and future of humanoid robot technology [BNP Paribas Asset Management]
Humanoid robot: The AI accelerant [Goldman Sachs]
The CEO’s Guide to Physical AI [BCG]
Exclusive: Founded By Ex-Nvidia Researchers, Flexion Lands $50M To Build The ‘Brain’ for Humanoid Robots [Crunchbase]
Flexion Raises $50M to Build the Brain of Humanoid Robots [Flexion]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Tech Week Takeover (NYC/BOS)

Thu, 14 May 2026 15:00:00 +0000

Hi {{first_name | there}} 👋

It’s that time of year again. Boston & NY Tech Week are just around the corner.

Imagine this: firesides with top startup founders, omakase dinners, and late-night afterparties.

The best part? You don’t need to imagine. When we promise a Tech Week takeover, we deliver a Tech Week takeover.

Consider this a special edition announcing not one, not two… but eight events Unicorner is hosting over the span of two weeks.

It’s going to be awesome. We’ll see you there. 🚀

P.S. Stay until the end for the greatest afterparty experience Unicorner has ever pulled off.

Arek and Ethan 🦄

The best way to learn how to build is from those in the trenches.

Excited to announce the Unicorner x Bolt x AWS AI Day, a full day + night experience at NY Tech Week built around the people and products shaping the future.

First, register and join us for three firesides with founders and execs at some of the most talked about companies in tech right now:

We’ll then transition into a private, intimate 30-person dinner for product, design, and engineering leaders at top companies who are looking to champion AI in their workplace (more on that below).

And finally, we’ll cap off the night with a live performance by Ian Asher. Guaranteed to be your favorite event of the week.

AI is reshaping how the best products are built. Join us for a private dinner with the leaders driving that change.

We’re bringing together 20 AI Champions at high-growth companies for a curated, three-course seated dinner. Expect sharp insights, great food, and an honest conversation about where AI fits into product strategy.

This dinner will be especially relevant to:

Heads of Product, Design, or Engineering
Lead/Staff PMs, Designers, or Engineers
Founders at Series A+ companies

Invites are extremely limited, and guests must confirm attendance to participate.

May 27, 6pm → RSVP here

June 3, 6pm → RSVP here

What’s better than one omakase dinner? Two omakase dinners. 🍣

We’re calling all emerging managers and founding general partners to join us for a small and highly curated omakase dinner: once in Boston, and once in NYC.

Spots are very limited, so if you’d like to join, RSVP below as soon as possible.

May 26, 6pm → RSVP here

June 2, 6pm → RSVP here

You’ve been to the panels, the rooftop happy hours, the accelerator mixers.

This isn’t that.

SF Tech Week Afterparty, October 2025

Bolt, Unicorner, and Flite are taking over the PUBLIC hotel’s rooftop for a private after-hours experience. Invite-only, approval required.

We’re closing out Tech Week with a live performance by Ian Asher. 11M monthly listeners. The man behind 3 billion plays in a single summer. Collabs with Tate McRae, Ava Max, Oliver Heldens.

The time for work is over. Time to unwind.

We’ll see you there.

This events is 21+, and IDs will be checked at the door.

We’ve got a lot going on, and we hope to see you there.

Questions? Comments? Event ideas? Reply to this email, and let us know. 🦄

Inception

Mon, 11 May 2026 15:00:00 +0000

Happy Monday.

Heading to Boston or NYC at the end of this month? We have eight events we’re hosting - be sure to register if you’ll be in town.

Tues. May 26 - Founding GPs VC Omakase (Boston)
Wed. May 27 - AI Champions Dinner (Boston)
Tues. June 2 - Founding GPs VC Omakase (New York)
Wed. June 3 - AI Day (New York)
Tues. June 23 - AI Champions Dinner (San Francisco)

Arek and Ethan 🦄

When you ask a traditional LLM like ChatGPT to write something, it builds sentences one word at a time using an approach called autoregression. The model represents each piece of a word as a token and generates additional tokens for each subsequent piece of the sentence it predicts. Meanwhile, image and video generation models rely on a process called diffusion, which creates an output by predicting the final outcome from a noisy input. Inception fuses these two approaches. Its flagship model, Mercury, generates text by starting from a noisy representation of the desired output and iteratively denoising it, producing the complete response in parallel.

An example diffusion process. (Structures Blog/Peter Sorrenson)

By generating multiple tokens at once, this method reduces latency and computational cost compared with token‑by‑token generation. The platform also supports multimodal integration, allowing language to be combined with audio, images, and video. It is currently deployed for enterprise customers.

Check it out: inceptionlabs.ai

Your competitors just shipped something. Probably with Bolt.new.

Founders are using prompt coding to launch products in hours, not sprints. Describe your app, Bolt.new builds it. Landing pages, dashboards, and full-stack tools.

No hiring. No handoffs. No six-week timelines.

The founders moving fastest right now aren't writing code. They're writing prompts.

Start building at bolt.new.

This is sponsored content.

Inception’s business model centers on enterprise API sales. Its core offering is an API product tailored for enterprises running LLMs in production, targeting application-layer companies building AI features. Key use cases include applications with strict latency requirements, such as voice agents, search and retrieval workflows (query writing, ranking, and summarization), and autocomplete and edit suggestions for coding tools.

Raised $50 million seed round in November 2025, led by Menlo Ventures with participation from Mayfield, Innovation Endeavors, Microsoft’s M12 fund, and Nvidia’s venture
Deployed in several Fortune 500 companies including Quora, Microsoft, and AWS

CEO Stefano Ermon’s lab at Stanford has been pioneering the research and development of diffusion models since 2019. Motivated by the question of why diffusion models worked so well for images and video but not yet for text or code, he and his team achieved a breakthrough in 2024 with discrete diffusion models for language, showing comparable performance to autoregressive models but 5 to 10x faster. This finding resulted in “Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution”, a paper that won a Best Paper Award at the International Conference on Machine Learning in 2024.

Ermon began exploring whether this approach could scale to production-level language models. He then reached out to investors and, together with two former students, UCLA professor Aditya Grover and Cornell Tech professor Volodymyr Kuleshov, founded Inception to commercialize diffusion-based LLMs.

Inception Lab’s team is undoubtedly exceptional. It’s hard to imagine someone better suited to start a diffusion-based LLM company than one of the creators of the diffusion model himself. But as Mercury 2 rolls out, two questions remain: do customers care enough to pay, and can diffusion-based models become a durable business model?

On the demand side, Mercury’s main advantage is speed. Inception claims Mercury 2 can reach 1,009 tokens per second on NVIDIA Blackwell GPUs, with pricing set at $0.25 per million input tokens and $0.75 per million output tokens. Launch coverage reports Mercury 2 achieves around 1,000 tokens per second, compared to just 89 tokens per second for Claude 4.5 Haiku Reasoning and 71 tokens per second for GPT-5 Mini, making Mercury 2 roughly 11 to 14 times faster in these comparisons. Independent benchmarks show slightly lower, but still competitive numbers, with Mercury 2 at about 873 tokens per second on leaderboards and 672 tokens per second in other tests depending on hardware and setup.

Speed alone wouldn't matter if quality suffered, but it doesn't. Mercury 2 posts 91.1 on AIME 2025 (competitive math), 73.6 on GPQA (graduate-level science), and 71.3 on IFBench (instruction following), which is in the same tier as Haiku and GPT-5 Mini.

Diffusion-based AI models optimized for coding, real-time voice, and instant agents.

That combination is what makes the value proposition real for latency-sensitive products, and early traction supports it. Mercury is live on Azure AI Foundry and is powering Microsoft's NLWeb. Thus, the path to market victory for Inception hinges on selling to customers who value speed at Haiku-tier quality over the intelligence gains of frontier models like Opus 4.7, and on competing primarily with speed-optimized peers like Haiku rather than frontier reasoners.

Durability is the tougher challenge for Inception. Diffusion offers architectural benefits including parallel generation, built-in error correction, and strong support for multimodal tasks, and the founding team is world-class experts in this field. While major players like OpenAI, Anthropic, and Google (which has already published its own research on diffusion language models) will quickly adopt diffusion if it proves successful, Inception currently leads.

The growing appetite for low-latency inference is apparent across the industry, as other companies are tackling latency from different angles, such as Groq and Cerebras, which are optimizing hardware for autoregressive models.

Whether diffusion becomes a lasting category and a true alternative to autoregressive models, or simply a feature absorbed by incumbents, remains to be seen. Recently, only a few dominant players have shaped the AI landscape, but investors are now betting that Inception could join their ranks. If Inception can demonstrate market viability and drive adoption of its diffusion LLM, it could become a household name.

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

The $10 billion idea everyone ignored

Thu, 07 May 2026 15:00:00 +0000

It was a Saturday morning, and Maya had finally unpacked. After the blur of boxes and the muttered I'll never buy this many books again, the apartment was starting to feel like hers.

She grabbed her keys and headed out to meet the new neighborhood. Her friends had insisted it was worth the price, because it had subway access, good restaurants, a farmers' market on weekends, and a building on the Bilt network. She'd signed up when she accepted the lease, almost without thinking about it.

Now, every month when she paid rent on time, she earned points. These points were also earned at the coffee shop she was walking toward, the Walgreens down the street, the Soul Cycle studio she kept meaning to try, or saved up for a flight or hotel stay she'd been putting off.

Her last apartment had given her nothing but a receipt. This one was giving Maya a neighborhood and helping her build a credit score that, somewhere down the line, would help her own a place.

It seems so simple now, but getting there took nearly a decade, a regulatory battle in Washington, two years of pitching and receiving nos, as well as one stubborn founder who refused to accept that the biggest expense in a generation's life should count for nothing.

This is how Bilt got built.

Welcome back.

Every other Thursday, Zero to Unicorn shares the life stories of billion-dollar founders, investors, and operators.

Join us at our upcoming events.

Tues. May 26 - Founding GPs VC Omakase (Boston)
Wed. May 27 - AI Champions Dinner (Boston)
Tues. June 2 - Founding GPs VC Omakase (New York)
Wed. June 3 - AI Day (New York)
Tues. June 23 - AI Champions Dinner (San Francisco)

Ethan and Arek 🦄

Today, Bilt is a $10 billion company with a network of more than 50,000 merchants. It’s founded by Ankur Jain, who is a World Economic Forum Young Global Leader and a Wharton School of Business at the University of Pennsylvania graduate. He was also named by Inc. magazine as the "Best Connected 21-Year-Old in the World". And, he has accomplished all this at a time when most people around him were still trying to figure out what they wanted to do with their lives. But to understand how he got there, let’s rewind time and begin with his parents.

Young Ankur featured in Forbes 30 Under 30. Courtesy of Bilt.

Ankur’s father, Naveen Jain, came from a poor village in rural India and earned his way into the Indian Institute of Technology, one of the most prestigious academic institutions in the world. There, the PC revolution taking place gave him a path to something bigger. In 1983, he came to the U.S. on a business exchange program.

This same relentless will to turn ideas into reality was a constant in his life. It showed up again, maybe most memorably, right after his wedding. Ankur’s mother, Anu Jain, had grown up in Israel, and neither she nor Naveen could have imagined where they'd end up spending their first days as a married couple.

That’s because, with no money for a honeymoon, his father had a… how can we put it? A singular idea.

The logic was, if he sent his resume to twenty companies in cities across the U.S., and even one of them flew him out for an interview, he'd have his trip destination.

Long story short, in 1989, a company in Seattle called Microsoft (already one of the fastest-growing tech businesses in the country) responded. He went for the interview, and the couple had their honeymoon in Seattle. That's where Ankur was born. And what started as a honeymoon turned into a career, because Naveen would stay at Microsoft for nearly seven years, eventually rising to group manager of MSN.

This figure-it-out mentality that brought Naveen to the U.S. would drive him to found InfoSpace in 1996, one of the pioneering companies of the early internet era, and later to launch ventures in space and health, such as MoonExpress and Viome. He became the kind of entrepreneur who goes looking for problems that nobody else has tried to solve properly.

Like father, like son, they say. That’s because it's hard to outrun the household you grew up in. For Ankur, this meant inheriting an instinct to solve problems with anything within reach, not waiting for the ideal context. And that’s what he did at an early age.

Ankur and his father in 2024.

Ankur grew up in an immigrant-parents household where asking to buy something was, by his own description, like “triggering an allergic reaction”. So, if he wanted anything, he had to figure out how to build or barter his way to it. Around 11 years old, convinced by a late-night TV ad that a pair of $140 plyometric jump shoes would let him dunk a basketball and make the school team, he built a website to get those shoes.

It had quizzes, AIM profile tools… to him, it was enough of a product to cold-call the company's customer service line and ask to speak to the CMO. He kept calling until someone actually put him through. There, he landed his first ever deal: an ad placement on his site in exchange for a free pair of shoes. The shoes did not make him dunk, but the deal-making instinct and the reflex to find a way through any problem stuck.

Clearly, it didn't come from nowhere. The fire that Ankur’s father carried from a small village in India all the way to a Microsoft office in Seattle never really went out. That serial entrepreneurship passion for solving problems would be imprinted on his children, all of whom would become entrepreneurs and important figures, and that’s how Ankur would build companies of his own.

But before that, he'd spend two years working for someone else's company, something he had never done before and would never do again.

You're at a rooftop bar in Austin. You meet someone, exchange numbers, and maybe promise to grab coffee. Life moves on. Three months later, their name pops up on your phone, and you have no idea who they are. Is it an old friend from school? The friend of a friend who could help you land a new job? Ankur knew that feeling, and it was one of the first problems he tried to solve by co-founding Humin right out of Wharton.

Humin was a contacts app he describes as a better version of LinkedIn, built around your phonebook, not a website. It helps you remember people by cross-referencing your calendar, social apps, and location to give you contextual awareness of the people around you. But the product never quite found its footing with a mainstream audience, and after nearly four years, Tinder came knocking.

The seeds for Humin’s acquisition by IAC, the parent company of both Tinder and Match Group, were planted at Web Summit in Dublin in November 2015, where both Ankur and Tinder's co-founder and CEO at the time, Sean Rad, were speakers. The two started discussing how to help people build real-world relationships, and what began as a conversation about potential partnerships became a full acquisition. Humin's intellectual property was folded into Tinder's product roadmap, with Ankur joining as VP of Product in 2016, the only time in his career he would work for someone else.

Almost two years in, he'd helped turn Tinder into the largest dating app in the world, and by his own account, it was great.

"It was an amazing experience, (...) it was fun. You just create products that change people's behavior in real life and help bring millions of people together.”
Ankur Jain

But there's only so much optimization you can do with a swipe left-right platform before the bigger questions start getting louder. He said himself, "It was just time to go back to my roots. How do you solve problems?"

The answer was waiting for him in SF, in an unexpected form.

Just after Ankur left Tinder in 2017, a chance encounter made the bigger questions ring louder. Freshly back in SF, he sat down with an investor who, with complete sincerity, pitched a company he believed would change the world.

It was… a startup that had figured out how to put purses on the blockchain so people could own unique digital handbags.

This bothered Ankur enough that he couldn't shake it. In his opinion, he saw founders pitching $100 million bets on digital products while real crises such as housing, healthcare, and student debt go largely ignored. He shares how SF is a bubble that he strongly dislikes:

“When everybody's in the same industry, there is such an echo chamber that gets created. It's like the worst of social media combined into one, like a microcosm of a city… everybody's talking points, all you hear about is the same thing. And you totally lose touch with the rest of the world”.
Ankur Jain

His answer to break this self-absorbed microcosm was to build a structure for attacking real problems systematically. The seeds of it had been planted years earlier, back in 2008, when he was still a freshman at Wharton, and the financial crisis hit. Overnight, the seniors around him lost their banking and consulting offers.

He looked around at all these smart, suddenly jobless people across campuses at the same time that a college kid could the next big thing in a dorm room. There was Facebook taking off, the iPhone launching a year prior in 2007…

That realization gave birth to The Kairos Society, a non-profit he founded to do what Y Combinator was doing before YC was really a thing, but for university startups around the world.

He'd always believed one thing above all else:

"If you see a problem that you get stuck on and you can't stop thinking about it… that is the company you should start."
Ankur Jain

Kairos was his attempt to find other talented people who thought the same way.

It was a fellowship for some of the brightest students from top universities around the world who believed business could be a force for solving real problems, such as energy, education, and health.

What started as a tight-knit community grew into a global network spanning 55 countries, connecting the young founders with leaders working on the world's biggest issues through events and direct mentorship. He was the founder and chairman of this organization until 2012, when he stepped down to focus on building Humin.

A newspaper article about Kairos. Courtesy of Bilt.

But by 2017, with a clearer sense of what he wanted to build and why, he reworked it into what is now Kairos HQ, a studio and family of brands united by the diagnosis that an entire generation had been set up to fail.

The company shares how this is the first generation projected to do worse than their parents financially, graduating with mountains of debt, spending nearly half their income on rent, unable to save nor buy, stuck in what Kairos calls “financial purgatory.” The studio's entire reason for existing is to find the places where the system has broken down most visibly for young people and build something better in its place.

As CEO of Kairos, Ankur doesn’t found each company himself. Instead, he brings together entrepreneurs who share that diagnosis and gives them the structure, resources, and network to go after their corner of the problem.

There’s Rhino, which wants to replace costly security deposits with a low-cost insurance policy, saving renters thousands on move-in costs. Little Spoon aims to make it easier and more affordable for parents to feed their kids well. Then, there’s Alloy, which focuses on helping women going through menopause manage their symptoms. Last, but not least, Cera’s goal is to bring quality home care to aging parents.

An article about Rhino from 2019. Courtesy of Bilt.

Of all the problems he kept seeing with Kairos, one kept coming back.

Ankur saw, over and over again, young adults spending a large chunk of their income to pay rent on time every month, but they had no reward whatsoever for this large commitment.

The more he saw it, the less he could unsee it. So he did what he'd always said you should do when a problem won't leave you alone.

He built a company to solve it.

That’s the first spark of Bilt, incubated at Kairos.

For most millennials and Gen Z, payday has a bitter ritual. Two minutes of happiness staring at your balance and then, dollar by dollar, it disappears into the usual suspects, like car payments, student loans, health insurance, utilities, and streaming subscriptions.

But the one that hurts the most, month after month, is rent.

It’s probably the most reliable financial commitment of your adult life. And yet, when you walk into a bank and try to buy a home, they look at your file and tell you that you don't have enough credit history. There’s no acknowledgment that you've been handing over a third of your paycheck without missing a beat for years.

"If you pay your rent on time for five years and go to buy a home, they shouldn't say you don't have credit history. That's insane," Ankur would say.

So, he was going to be the one to fix it.

The idea was simple. You pay rent every month on time, so why doesn't it work like every other expense in your life? If you buy a coffee, you earn points. If you book a flight, you earn miles. But somehow you pay your biggest bill, and get absolutely nothing. Over 21 million renter households in the U.S. spend more than 30% of their income on housing, and at the end of it all, there's no reward.

The original Bilt pitch from 2018 asked, why can't paying rent work like a frequent flyer program? You pay rent on time, earn points that could later be used toward a home down payment, while also building the credit history that would eventually help you buy such a home. After all, if you’ve been paying rent on time for the last five years, shouldn’t that serve as proof of your finances? The idea was to turn renting from a financial dead end into a path to homeownership.

The inspiration for the whole model actually came from a conversation with Barry Sternlicht, chairman of Starwood Capital Group. Barry had told him that airlines and hotels make more profit from their loyalty programs than from flying planes or operating hotels.

Ankur thought that if that model could be applied to housing, the opportunity was enormous. Barry believed it, too, enough to eventually become the first landlord to sign on, as we will soon see.

For the moment, Ankur had to roll up his sleeves and get to work to make this idea come to life.

But he hadn't even properly started when the first "no" arrived. And, it threatened to sink everything before it even began, because it came from the law itself.

Six months after starting the company around this approach, Ankur received a call from his lawyers.

The good news? They loved the idea. The bad news? They weren't sure it was actually legal.

It turns out that post-2008 financial regulations had tightened the rules around what could be used toward a home purchase, and when regulators wrote those guidelines, nobody had considered rewards points. So it wasn’t explicitly permitted… but it wasn’t prohibited, either.

They had two choices. Either build anyway and hope it would never cause any trouble, or fight to get it properly approved.

Ankur chose the harder path, partly out of principle and partly because, without venture money, he wasn't in a rush (yet). That was only possible because of the Humin exit, which had given him a financial cushion, and he was determined to use it wisely, without answering to anyone.

That instinct went all the way back to the kid who built a website to barter for a pair of shoes.

In an interview with Forbes CEO Mike Federle, he'd later put it plainly:

"This is like the type of hustle that we forget about. How do you get really creative with limited resources to solve problems? And this is another reason why raising too much money too early is dangerous. Because no matter how smart you are, you start solving problems with money first, versus how do you get creative with a win-win-win to make things work?"
Ankur Jain

It comes to the idea of money being as much a trap as a tool.

When you raise venture money early, he would say, you're immediately in a race to show numbers you don't have yet, a kind of ticking time bomb that starts the moment you take it. Ankur wanted the freedom to get it right from the start. He intended to raise eventually, sure, but only from people who had skin in the game for the same reasons he did.

So he spent 18 months up and down from NY to Washington, D.C. meeting with regulators, the Secretary of Housing, and cabinet members at the White House, walking each of them through the same core argument: if Americans can earn rewards at a bar, on a flight, or at a restaurant and use those points to book a hotel room, why on Earth couldn't they use them to build a credit score and toward buying a home?

After 18 months, it finally seemed to have clicked. Regulators told him to submit a formal approval request.

A week later, a rejection letter arrived in his inbox.

That night, he panicked. His relationships came to the rescue.

Years of building Kairos and gathering people around real problems had left him with a network deep enough to call at midnight. By the next morning, he was in front of a full committee, appealing the decision in person. It was a reminder that the network you build before you need it is usually the one that saves you.

This time, it landed. In October 2019, the Federal Housing Administration formally agreed that paying rent on time could now qualify Americans for a home mortgage, and reward points could be used toward a down payment.

Ankur on the left, Ben Carson in the center, and Rick Perry on the right. Courtesy of Bilt.

So, what started as a simple question now had a legal answer. Members could earn rewards, build credit, and work toward a lower mortgage rate, all from the bill they were already paying every month.

With the law on his side, you’d think the biggest obstacle was behind him.

Unfortunately, that was just the beginning.

They say when one door closes, another one opens. Sometimes that's true, but sometimes everything is closed, and not a damn thing opens, so you have to go there and break it down.

Stay tuned for Part II, where a global pandemic, of all things, helped turn things around. Clearly, it worked. Better than anyone expected.

Stay tuned.

Sphinx

Mon, 04 May 2026 15:00:00 +0000

Happy Monday.

We’re ramping up for Boston and NY Tech Week. But first, we have an announcement to make.

We just launched a new dinner series: AI Champions. We’re hosting intimate, curated dinners for product/engineering/design executives, senior ICs, and founders who are championing AI at their organizations.

Our first dinner (in partnership with bolt.new and MiniMax) was on Thursday. We’re hosting in Boston and NYC, as well as SF again.

Want to join? All registration links are below.

Eric Simons, Bolt’s CEO, presenting at Thursday’s dinner

Tues. May 26 - Founding GPs VC Omakase (Boston)
Wed. May 27 - AI Champions Dinner (Boston)
Tues. June 2 - Founding GPs VC Omakase (New York)
Wed. June 3 - AI Day (New York)
Tues. June 23 - AI Champions Dinner (San Francisco)

Ethan and Arek 🦄

Sphinx builds AI systems that perform compliance work inside financial institutions. It focuses on areas like anti-money laundering, know your customer, and know your business, which are still handled largely by human analysts.

In most financial institutions, this work is done case by case. Analysts review alerts, gather information across systems, make a judgment, and document the outcome for internal teams and regulators. As transaction volume grows, the process becomes slower, more expensive, and harder to manage.

Sphinx replaces that workflow with a system that handles cases directly. It reviews the input, pulls the relevant context, reaches a decision, and produces a record that can be audited later. The product can run end to end or in a human-in-the-loop setup, depending on how much control the customer wants to keep.

Check it out: sphinxhq.com

Sphinx sells through a demo-led enterprise process. It offers a free data test on five to 10 real case samples before any commitment. In production, customers are seeing up to 4x cost reduction compared with its previous human-led approach.

Raised $7.1 million seed round in February 2026, led by Cherry Ventures
Took part in the Y Combinator F24 batch

A Series D fintech automated more than 250 hours of manual compliance work per month without adding headcount
One client cleared a 12,000-alert backlog in a single week
Customers report up to 10x faster resolution times, and Sphinx says it is automating up to 92% of compliance operations in production
Equals Money, a London-based fintech, went live in a week with zero engineering work
Axos Bank, a digital bank born in 2000, is deploying Sphinx across its compliance organization

Alexandre Berkovic and Chrisjan Wust met at a braai, a South African barbecue, at OceanHub Africa, supporting African-led ocean-impact startups, and ended up living together for a short period in 2021. Two years later, they ran into each other again in a small town near Lisbon, Portugal. That second meeting led to Adorno AI, a generative AI audio and video post-production company they built together.

After about a year, the duo recognized that market circumstances weren't in their favor, and Adorno wouldn't scale into a billion-dollar business. So they made the pragmatic call to find a clean acquisition path and channel their energy into something new. The next company came from a problem Wust already knew well. As employee No. 1 and head of engineering at RelyComply, a risk management and compliance company, he had spent more than four years watching compliance teams work through slow, manual processes with systems that felt far behind the scale of the problem.

The gap had become harder to ignore. Analysts were spending weeks on cases that should have taken minutes, while fraudsters were using AI-generated documents, and legacy systems were still missing obvious signals. Berkovic brought a complementary lens from his background in Design Engineering at Imperial College London and machine learning at MIT Sloan. In 2024, together, they created Sphinx, pivoting from audio to compliance, and three weeks later, were accepted into Y Combinator.

More transactions create more alerts, more reviews, and more manual work. Compliance scales with volume, and, in most institutions, that still translates into additional analysts, with little improvement in fiscal or operational efficiency over time.

The root cause is poor data quality and outdated rule-based systems, which generate over 100 million false alerts per year in the U.S. alone, according to Sphinx. An estimated 95% of all AML alerts are false positives, meaning the systems designed to catch suspicious activity are almost never flagging actual threats. At scale, that inefficiency adds up to a massive operational burden. Global AML compliance costs are estimated at over $274 billion annually, much of it spent managing low-quality alerts rather than catching criminals.

That is the system most compliance teams are still operating in today, relying on analysts to move cases across disconnected systems, gather evidence, and document decisions. Traditional RegTech, or compliance software, improved visibility. Outsourcing added capacity. Neither changed how the work actually gets done.

Compliance flow for collecting documents, identifying UBOs, and requesting additional verification data.

Sphinx is building AI compliance analysts that work through cases, produce decisions, and leave behind an audit trail that can hold up under review. It operates as the analyst itself inside existing systems. That is a much bigger claim than workflow assistance but it is also where the upside is.

It stops scaling compliance in a straight line with volume, reducing the need to add analysts as they grow. The early production numbers point in that direction. At Alviere, an embedded financial products provider, Sphinx achieved 86.1% false positive detection and 88% automatic case closure, saving about 17 days of analyst work in a single month. That is the shift: less manual review and faster resolution without adding headcount.

The category is still forming, but it will not stay that way for long. Incumbents will add AI features, newer companies will attack pieces of the workflow, and internal teams will try to assemble their own versions. But most of those approaches will still be centered around the analyst. Sphinx is more ambitious because it is trying to replace the work itself.

Compliance has always scaled linearly with volume, with more activity requiring more analysts. That model is starting to break as fraud accelerates (with AI-generated docs, synthetic identities, deepfakes, and so on), transaction volumes expand (think about high-throughput blockchains, agentic commerce, and tokenization of real-world assets), and regulatory pressure increases. The institutions that adapt will be the ones that can break that relationship.

Sphinx Raises $7.1M to Build Every Financial Institution’s Last Compliance Hire [Business Wire]
Y Combinator Profile [Y Combinator]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Prospero

Mon, 27 Apr 2026 15:00:00 +0000

Last Thursday, we gathered in SF for a candid fireside with Alvin Salehi, a 2x YC alum, Harvard Law grad, ex-White House tech advisor, and co-founder of Shef, the delivery marketplace for home-cooked meals.

It was an amazing evening alongside the Auth0 and Fidelity Private Shares teams. Great people, awesome conversation. Exactly the kind of room we love.

Alvin Salehi (right) and Ethan Keshishian (left) sharing insights.

If you want a seat at the next one, registrations are still open for our upcoming events below.

AI Champions dinner, April 29 (SF): In partnership with Bolt.new and MiniMax. If you’re spearheading AI usage at your company, this dinner is for you. Apply to join us here.
Seed+ founders dinner, April 30 (SF): An intimate founder dinner, cohosted with Fidelity Private Shares. Invites are extremely limited; apply here.

Arek and Ethan 🦄

Prospero AI aims to give retail and everyday investors the ability to understand stock signals like someone on Wall Street. The Prospero app combines hedge fund experience with 10,000+ AI models and 100,000,000+ data points to turn complex market data into clear signals, helping investors make more confident decisions. Prospero believes that real-time market intelligence, important datasets, and stock signals have either been gatekept within the finance industry or remain too complex for everyday investors to interpret. This is why it aims to demystify higher-level finance with insights into where hedge funds and institutions are putting their money, while breaking down the analysis and the why. Rather than just giving stock picks, Prospero focuses on educating users so they can make informed financial decisions.

Check it out: prospero.ai

Prospero’s app provides users with information around stock signals and analysis powered by AI, along with stock screeners and picks built on data typically less accessible to retail traders. The team is continuing to develop and add premium features within the app, such as notifications about changes to specific signals for given stocks. Prospero also has a newsletter with both a free and a premium tier.

Crowdfunded $538 thousand in 2025 on Republic (previously covered by Unicorner) with more than 500 investors
Grew from 10 thousand to 20 thousand users after its crowdfunding round, all with no marketing spend, and currently generating $100 thousand in ARR
Has 12 thousand to 15 thousand active weekly newsletter readers
Named Best AI Fintech Start Up USA 2025 by Global Financial Market Review
Raised over $1 million in angel funding

If you’d like to explore opportunities with Prospero, let us know by replying directly to this email!

Prospero founder and CEO George Kailas started value investing when he was just 17 years old. In simple terms, that means looking for companies priced below what they’re worth, then profiting when the market eventually catches up. From an early age, he gained hands-on experience with investing, but felt there was a lack of clarity around how it worked. While big, notable finance blogs had stock picks, there wasn’t a thorough and beginner-friendly explanation for the everyday retail trader. This often led them to blindly follow picks without ever understanding why.

Kailas continued to develop expertise, working as an analyst/investor at several companies. As someone who was both passionate and an expert, he wanted to provide similar access and explainability to everyday investors, which spurred him to start Prospero in 2018. Over time, a growing fascination with modeling and AI pushed the project further, eventually expanding into an app and a premium newsletter.

Historically, financial institutions have held a significant edge in both the data they access and the insights they act on that most retail traders simply don't have. Prospero aims to change this, with its app and ML-powered insights built to democratize financial insights.

Per Prospero CEO George Kailas, while financial numbers are publicly available, they often can seem intimidating or gatekept, leading to a lack of understanding for retail investors. For instance, anybody can look at what Google’s price-to-earnings ratio is, but can they actually explain what a company’s P/E ratio means given its unique circumstances and how that informs their investment strategy? There are hundreds of numbers and ratios just like this one, and a lot of them can be noisy for people who are not working on Wall Street.

That’s why Prospero tries to empower users through investing literacy in addition to providing stock picks through its AI models. The app aims to share and educate everyday investors on short- and long-term signals, giving them the tools to filter stocks and better understand what certain figures and information truly mean. Rather than just blindly following stock picks, this allows users to think for themselves and arrive at their own conclusions.

Signals are designed to be easily interpreted but also technically rigorous. For example, “Market Similarity” is a long-term signal updated daily, reflecting how closely a given stock mirrors the performance of the S&P 500, while “Net Options Sentiment” is a short-term signal that indicates whether institutions are betting on a given stock to fall or rise and by how much in the shorter term. While users don’t need to know how these signals are calculated, the Prospero app helps retail investors build their own logic and analysis around each one.

Prospero uses AI and ML to make complex stock signals understandable for everyday investors.

Prospero’s newsletter is an extension of the app, with in-depth analysis and discussions about current market news and information. While plenty of news publications discuss finance, Prospero ‘s goal is to resonate with everyday investors through language for beginners, unlike other publications that skew toward those working in finance.

Kailas told us he was thinking about the business in two phases. First is trust-building, and second is monetization. He sees Prospero as nearing the conclusion of the trust-building phase, primarily driven by its content, and this has focused on the development of an audience/community through channels like the newsletter and Discord. Beyond its educational content, many of its stock picks and breakdowns have served as a way of establishing trust, with its picks outperforming the S&P 500 by nearly 13x.

Next is monetization. Prospero plans on adding more in-app features and new signals, focusing on an expansion into wealth management and launching its own ETF. Given that the team's resources and stock picks have established credibility with its community, it’s on the right track to monetize effectively.

While AI-enabled finance and investing content is also crowded, three strengths set Prospero apart. First, it knows exactly who it is serving (i.e., the retail investor). Everything else trickles down from that. Second, in a space with a lot of self-proclaimed experts, the company has an experienced and passionate leader who has been in the trenches as a real SME. Lastly, there is a strong community behind Prospero, seen through the app, newsletter, and Discord. If it can effectively grow its monetization while maintaining credibility among its target users, we can see Prospero being a go-to for investment education and selection.

Official Prospero Newsletter [Prospero Substack]
Prospero Crowdfund on Republic [Republic]
7 Best AI Investing Tools That Actually Beat the Market [Shares Prediction]
Why Prospero is the Future of Stock Screeners [Prospero AI Blog]
Retail investors are no longer following the market [Fast Company post written by George Kailas]

From bankrupt to $20 million ARR in 2 months, Part II

Thu, 23 Apr 2026 15:00:00 +0000

He’d taught himself to code.

He’d moved to Silicon Valley at 18 years old to pursue his dreams.

He’d founded not one, not two, but three companies.

But now, after seven years, Eric Simons was stumped. Despite everything, StackBlitz was on the brink of bankruptcy. And with a baby and an Ironman coming up, the stakes couldn’t be higher.

Eric was determined. In his own words, in fact, he’d never been happier.

Still, the team still needed a miracle.

With nothing left to lose, for all of 2024, they experimented.

Bolt was the last experiment on the list before the end of the year, and with it, StackBlitz.

Missed Part 1? Read the full article on our website.

Welcome back!

ICYMI: Every other Thursday, Zero to Unicorn shares the life stories of billion-dollar founders, investors, and operators.

They’re the untold tales of the miraculous founders who have made it to the pinnacle of their industries against all odds. They’re the impossible pivots that arose out of necessity, the years where the product was great but the business wasn't, and the moments our protagonist almost walked away. And expect the highly coveted, hard-won lessons they learned along the way.

We also host regular, in-person Zero to Unicorn founder fireside chats in SF and NYC. In fact, we’re hosting one tonight with Alvin Salehi, co-founder of Shef. RSVP, come by, and say hello!

That’s all for now. Enjoy Part II of Eric’s story!

Ethan and Arek 🦄

The first iteration of Bolt actually dates back to February 2024 and was based on a problem Eric had experienced in his own life.

At the Unicorner Zero to Unicorn Founder Fireside, he shared how he tried to build a website for his own wedding in 2021. Partway through the project, he was distracted by other priorities and never finished it. He eventually gave up and switched to Squarespace, which he found frustrating to use. The vision of this product was to eliminate exactly that kind of friction.

The team constructed a prototype, a barebones AI tool that could turn a simple text prompt into a working application. However, it failed because the models just weren’t good enough back then to make the idea work. The apps and websites it created glitched and fell apart. The seamless experience they were after, the simplicity of writing what you want and getting a full website or app ready to use, was nowhere close to feasible.

Zooming out, we’re in a moment where a lot of companies are scrambling to add AI to their products. The StackBlitz team held off because they saw the technology in its early, unusable state and didn't see how it would improve the experience they wanted to create. So they waited.

In retrospect, this was the right move.

In May 2024, the team had their perspective flipped. It was Anthropic's Claude Sonnet 3.5, a model that, in Eric's words, "has changed software development, period. We got a kind of sneak peek of it, and we were like, ‘This is good, this is going to change everything’."

They pulled Bolt off the shelf and went all in. For all of July, August, and September, they were heads down.

With a company, a baby, and an Ironman on the line, the stakes were high.

And the first outcome couldn’t be better. With less than 6 months of training, Eric crossed the finish line of a full Ironman at the end of October 2024: it was a 2.4-mile swim, a 112-mile bike ride, and a full marathon, completed in 12 hours and 15 minutes.

Eric Simons during his Ironman. (Courtesy of Bolt)

As impressive as that is, the success they soon saw with Bolt somehow topped it.

Twitter tweet

The “launch strategy” for a product that would go on to reshape how software gets built was a single tweet that went viral.

One day after the post, Bolt had added 60,000 users. One of the engineers messaged the team in disbelief.

But Eric wasn’t fazed. He had already seen his share of launches. They get people excited for a few days before fading away. He thought, this one would follow the same pattern.

To his surprise, it didn’t.

It just kept growing and growing.

Bolt reached $4 million ARR in just four weeks, crossed $20 million in three months, and hit $40 million in five.

It’s one of the fastest-growing startups of all time by revenue velocity, with only ChatGPT faster.

Its patience, combined with seven years of StackBlitz infrastructure nobody else had, meant that when the right model arrived, they could move in 90 days while competitors were still figuring out their roadmap.

The team finally had a moment to breathe, albeit a brief one. They knew they had a hit product on their hands. After driving through hell, they had finally made it out the other side.

What they didn’t expect was just how fast the market they were building in was growing.

"It was kind of like a murder mystery," Eric said during a Forbes Talks episode. "What is going on? Who are these people using it? Why are they using it?"

The users weren't coming from just one place, and they weren't one type of person.

Somewhere in the middle of watching Bolt grow in ways founders couldn’t fully explain, something clicked. Users were coming from everywhere because the shift happening underneath Bolt wasn't really about Bolt, but a fundamental change in what software development could look like and who could participate in it. Bolt happened to be in the right place with the right execution when that shift began.

This is why the company could have a smash hit with a product scraped together in 90 days that was, in Eric’s words, like a stripped-down race car:

“Just to be clear, ... it was a true MVP, it's kind of like when you see race cars, where they strip out everything. There's no passenger seat. It's just that there's no leather interior. It's just the metal, and that's what Bolt v1 was.”
Eric Simons

The product offered what Eric describes as a dead-simple experience: a homepage with the chatbox, and another page with a chatbox next to the product you’re building.

As Eric had learned earlier when building StackBlitz, friction at the start kills adoption. Bolt from day one aimed to eliminate just that. Going from "I have an idea" to "I have a working app" in under a minute was the only metric that mattered, and that’s what the team succeeded in optimizing for.

It turns out that the two-page experience of the product got its result, with 500 to 1,000 support emails a day and a Discord community scaling faster than any team of 15 people could formally manage. This community quickly became one of the company’s most valuable assets, telling them what was broken and what mattered most.

The result was a company that had been one board meeting away from shutdown, now sitting at the center of an entirely new category it helped create.

Bolt is simple. Describe the app or website you want, and it builds it for you. It’s neither a template nor a drag-and-drop layout. We’re talking about a complete, functioning web application generated from a simple prompt. If something isn't right, you tell it, and it fixes it. The whole cycle that used to take weeks of back-and-forth with developers now takes minutes.

And what makes it different from tools that just generate code and hand it back to you? Bolt actually runs it and debugs it in real time, autonomously. You just have to prompt it. The combination of a capable AI model plus a real execution environment is what separates Bolt from tools that just generate code and hand it back to you.

So, as Bolt’s user base exploded, the team faced a strategic question. Who exactly should they build for? They realized that consumers and prosumers drive top-of-funnel growth and that mattered, but who would use the product in the long term?

Beyond one-time users and solo projects, Eric realized the customers who would open Bolt every single day were businesses and enterprises, because they were the ones with the power to transform how entire organizations build products. If Bolt wasn't good enough for them, it would become a toy. And that was another lesson that Eric had already learned.

It's also where most competitors aren't. Consumer-facing AI coding tools have multiplied rapidly, but building for businesses is a very different problem. It means dealing with security, integrations with existing systems, collaborative workflows that include non-technical teammates, and much more.

On top of that, agents were just starting to get good.

These shifts shaped the next phase of the product. In other words, it was time to supercharge their race car.

Think about the product manager who has a clear vision for a new feature. Under the old model, normally, they write a detailed memo, route it for approval, pass it by the design team, and get yet another round of approval. Then it lands in an engineering queue, where it waits.

Weeks, and sometimes months, pass before anything resembling their original idea comes to life, and by then, it's been through so many hands, it barely resembles what was originally imagined.

Eric has a name for all of that: “fake work.” It’s all the meetings, the approvals, the translation layers between an idea and its execution.

"There are a lot of brilliant people at companies who have been held back by arcane processes," he said during the Forbes Talks episode. And the people who feel that gap is broad, including product managers, designers, marketers, and founders.

With Bolt, the PM types what they want, a working version comes back, they show it to people, and it ships. It’s a way of giving agency to people again.

In October 2025, the second version of Bolt arrived, designed for a different kind of customer. It was the kind of product that 75% of the Fortune 500 now use.

If v1 was a stripped-down race car, v2 was the Ferrari. Containing fewer error loops, projects that could scale to a size that matched enterprise needs, and the infrastructure those companies required, including full-scale databases, authentication, payments, security, SEO, and so on, Bolt handled everything inside the platform without any setup.

And at the center of it all? Professional-grade AI agents. The broader industry is shifting from AI models (tools you query and get answers from) to AI agents. They are systems that go beyond natural language processing to autonomously handle decision-making, problem-solving, and direct interaction with external tools. These tools break down complex goals into smaller tasks and use past context to personalize and refine future interactions.

Last, but definitely not least, agents can work together. Multi-agent systems unlock the ability to execute complex, multi-step workflows by having specialized agents coordinate and run concurrently, which can take care of complex tasks that used to require entire teams. In other words, we’re talking about fewer bottlenecks and dramatically reduced operational costs. For anyone managing teams and budgets, that's hard to ignore.

Eric sees this as the defining transition of the next decade.

He believes every person at every company will eventually have their own agentic team. Bolt is oriented around this concept.

"One person with a team of 50 agents is going to be faster than 50 people in an office," he said. And that's not just a prediction for Eric, as it's already how Bolt operates. The company has paused traditional hiring, choosing instead to bring on agentic systems across engineering, product, design, marketing, and support.

He expects this entire model to become the norm across the industry, with smaller teams, multiplied agency, and building things at a pace that simply wasn't imaginable before.

The metric he uses to measure this is gross margin per employee. Most companies don't think about productivity this way, but Eric believes it is the clearest signal of how much leverage a single person has.

Bolt's target for the next 12 to 18 months is $10 million in gross margin per employee, a number that sounds almost implausible, and Eric knows it. But he believes it’s inevitable, and he wants Bolt to be the first ones to achieve it.

If Bolt keeps on this path, we have no doubt that it will.

The risk on the horizon is the frontier labs. Companies whose models power platforms like Bolt, like OpenAI and Anthropic, are not going to sit still. Eric has no illusions about this. Asked during Unicorner's Zero to Unicorn Founder Fireside whether they'll eventually build their own Bolt-like products, his answer is straightforward: "I have no doubt that they will."

But he doesn't think that closes the game. Bolt's answer to that threat is flexibility, the ability to let users reach for whichever agent or model is best for any given task, regardless of who built it.

Besides, Bolt isn't waiting around, as the numbers show. The company raised $105.5 million in a Series B, led by Emergence Capital and Google Ventures, in January 2025. The growth that followed was hard to ignore, with more than one million AI-generated websites built then deployed on Netlify alone by March 2025 and 10 million users by February 2026.

However, the number Eric is most focused on is where the product is being used day in, day out inside real organizations. Bolt has since 10x-ed its B2B business: as mentioned before, it is now trusted by three-quarters of the Fortune 500. It has also earned a spot in the top 40 fastest-growing software vendors globally, according to Brex.

The product has kept pace with the momentum, with new model integrations landing regularly, and Bolt Connectors bringing tools like Miro directly into the workflow, closing the gap between how teams design and how they build. The team also just launched Design System Agents, allowing users to generate code that is native to your design library, and again demonstrating an ability to stay ahead of the curve. On top of that, Eric recently hinted at what's next: a way to take ideas straight into production codebases without the usual rewrite cycle, something he says product teams have wanted for a long time. We won’t spoil the surprise, but we’re confident it’s worth following.

Behind all the funding rounds and model integrations, what matters most at Bolt is the experience for the user. At the heart of it, that experience proves to be both powerful enough for professionals while still simple enough to use that, yes, even a chemistry teacher who once asked a distracted teenager what would make him pay attention in class could build something with it. Funny how life turns out.

Bolt's story looks like an overnight success, but it was eight years in the making. The category Bolt helped to create didn't exist three years ago. It exists because one team sat at the intersection of the right technology, the right moment, and the quiet, unsexy work of not giving up before the pieces came together.

Your competitors are free R&D. When other companies experiment with AI features, new markets, or product bets, watch closely. Let them absorb the cost of figuring out what doesn't work.

“They're going and deploying that capital to go and do things, and if they're successful ... success leaves clues, and you can look at those, and you can … pull into your own strategy.”
Eric Simons

Don't spend faster than you need to. Before product-market fit, your investors will push you to scale. Resist it.

“You want to buy as much time as you can, because you have no idea how much time it is going to take.”
Eric Simons

When the 2023 ultimatum came, StackBlitz still had a year, maybe a year and a half of runway in the bank. So, the company wasn't at imminent risk of running out of money. The harder question keeping everyone up at night was: What are we doing here? That disciplined spending is what bought them enough time to find out.

Community is distribution. When Bolt launched, it wasn't starting from zero. Seven years of StackBlitz users became the first wave. And when the product caught, it caught hard. Recall the Discord community flooding the team with exactly who was using it and why.

“Once you have something that is great, you want to have some number of people that are following you that you can go and ... put in their hands and then it'll go viral and kind of catch like wildfire.”
Eric Simons

Know why you're not giving up. Persistence is good advice, until it isn't. The question to ask yourself is whether you're holding on because you have real conviction, you want to see that working, or because giving up feels like failure.

But if you decide that’s what you want, Eric has a message to you:

“The important part there is, when you're driving through hell, don't stop. If you're going to go and build a business, go and try all the way through and don't leave any stones unturned.”
Eric Simons

Build for professionals first. A tool that works for everyone has to first work for the most demanding users. If it can't hold up there, it becomes a toy. Remember how developers loved WebContainers, but wouldn't abandon the local environments they had spent years perfecting. It was cool, but not a long-term business product. When it came to Bolt, he wasn't going to make the same mistake.

Talk to your customers obsessively. Eric's formula for maximizing innovation is a simple, timeless message to close with:

“If you want to put innovation to the max, that's directly just proportional to how much time you're actually spending talking to the customers that you're looking to serve.”
Eric Simons

Logic

Mon, 20 Apr 2026 15:00:00 +0000

ICYMI: we have a special event coming up this Thursday.

Alvin Salehi, co-founder of Shef (the delivery marketplace for home-cooked meals), is joining us for a candid conversation on April 23rd.

Spots are limited. If you want to hear about all things founder insights, be sure to join us for our latest Zero to Unicorn IRL chat.

Zero to Unicorn fireside with Alvin Salehi, April 23 (SF): A deep dive with the co-founder of Shef, in partnership with Auth0. Apply to join us here.
AI Champions dinner, April 29 (SF): In partnership with Bolt.new and MiniMax. If you’re spearheading AI usage at your company, this dinner is for you. Apply to join us here.
Seed+ founders dinner, April 30 (SF): An intimate founder dinner, cohosted with Fidelity Private Shares. Invites are extremely limited; apply here.

Ethan and Arek 🦄

Logic is building autonomous agents for enterprise workflows. Its system lets users deploy agents that can handle tasks like research and operations by working with the tools companies use, whether that is Slack and Notion for internal coordination, customer platforms for support workflows, or engineering tools and internal databases for technical tasks. Unlike a chatbot or copilot that mainly answers questions or helps with one-off tasks, Logic’s agents are built to handle multi-step workflows and operate with less human input over time. The platform fits into the broader shift toward AI infrastructure that helps companies automate internal work and improve efficiency.

Check it out: logic.inc

One prompt gets you a working app.

Not a wireframe, not a prototype. A live product.

Founders are skipping the backlog entirely and shipping what they need, when they need it.

No dev team required. No waiting. Just results.

Start building at bolt.new.

This is sponsored content.

Logic uses a usage-based SaaS model with team and API pricing. Team plans range from a free trial to $20 and $200 per user per month, with higher tiers offering more executions, longer log retention, added security, and enterprise controls. Logic also offers teams a free trial with limited executions. API usage is priced separately at $0.05 per execution.

Raised a $4.3 million seed round in October 2025 from institutional investors, including Audacious Ventures, Founders’ Co-op, and Convoy co-founder Dan Lewis, along with individual backers such as Neo CEO Ali Partovi
Became generally available in December 2025, removing its waitlist and opening the platform to users
Had more than 2 million automations run at the time and has since shifted toward a stronger focus on engineering and technical users
Built a platform supporting 200+ AI agents with a team of around six people

If you’d like to explore opportunities with Logic, feel free to reach out to us by replying directly to this email!

Steve Krenzel’s path to founding Logic was shaped by experience across both large tech companies and startups. He started his career as a software engineer at Microsoft before co-founding Thinkfuse, a workplace communication startup that helped teams streamline internal status reporting. It was later acquired by Salesforce, where he met his future co-founder, Jess Garms. After Salesforce, Krenzel worked at Twitter and later joined companies including Convoy and Brex.

Across all of these roles, he kept seeing internal teams struggling with clunky tools and inefficient, hard-to-scale workflows. Logic came directly from that recurring problem. Instead of adding yet another tool for teams to manage, Krenzel wanted to build a platform where AI agents could carry out standard operating procedures on their own, including testing, execution, and recurring workflows that typically take a lot of human time. That idea became the foundation for Logic’s vision of helping small teams operate with far more leverage.

If a task takes more than 30 seconds, Logic believes it’s time for an AI agent to take over. This idea pushes AI beyond just assisting with work and into fully executing it, one of the biggest shifts in software today. McKinsey highlights how AI-driven development is transforming programming, moving from simple task-based support to AI agents running entire workflows and applications.

What makes Logic interesting is its focus on recurring work. Many AI tools still depend on people to prompt them at every step of the way. Logic is trying to move past that by building agents that can recognize recurring tasks and handle them more independently over time. That means the product becomes more valuable the longer it is used, building a moat for the company, and transforming it into a complete operating layer inside a company rather than just another productivity tool.

This vision is reflected in the product, too. Using Logic seems to be less about chatting with a single assistant and more about setting up a system for repeated execution. Users can turn standard operating procedures into agents that handle recurring tasks, run tests, and carry out workflows across the tools they already use. This makes the case for scalability very compelling.

Agents follow written specs, which define their behavior, inputs, and outputs.

The team itself is optimized for scalability, too. Logic has a team of around six people and already supports more than 200 agents. It suggests a future where smaller teams can get much more done without having to keep adding headcount, which creates the potential for a highly scalable business with strong revenue leverage.

There are still risks. The AI agent space is unquestionably getting crowded with agents becoming a core part of enterprise operations that could generate up to $450 billion in economic value by 2028. That kind of momentum will likely attract even more startups and larger incumbents into the category. Logic will need to stand out through product quality, reliability, and by minimizing adoption friction.

Still, the company’s thesis is strong. Work is clearly moving toward more automation and more leverage per employee. If Logic can keep improving its agents and deliver on its vision of more autonomous systems, it has a real shot at becoming an important company in enterprise AI infrastructure.

Logic website [Logic]
How to build an AI agent: from phototype to production [Logic]
Logic Raises $4.3M in Seed Funding [Finsmes]
Anthropic launches new push for enterprise agents with plugins for finance, engineering, and design [TechCrunch]
Create an Onboarding Plan for AI Agents [Harvard Business Review]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Flint

Mon, 13 Apr 2026 15:00:00 +0000

We love it when past Unicorner startup stories come full circle.

Flint CEO Michelle Lim is familiar to our team. In fact, she was employee #2 (and the first software engineer) at Warp, which we covered back in 2022!

Speaking of past Unicorner startups… we have an event to announce.

2x YC alumni. Harvard law graduate. Ex-technology advisor to the White House.

Alvin Salehi has done it all. And, of course, what we know him for best: co-founder of Shef, the delivery marketplace for home-cooked meals.

He's joining us for a candid convo in SF on Thursday, April 23.

Spots are limited. If you want to hear about all things founder insights, be sure to join us for our latest Zero to Unicorn IRL chat.

April 23 (SF): Zero to Unicorn fireside with Alvin Salehi, co-founder of Shef, in partnership with Auth0. Apply to join us here.
April 29 (SF): AI Champions dinner, in partnership with Bolt.new and MiniMax. If you’re spearheading AI usage at your company, this dinner is for you. Apply to join us here.
April 30 (SF): Seed+ founders dinner for 20 founders with Fidelity Private Shares. Invites are extremely limited; apply here.

Arek and Ethan 🦄

Flint allows marketing teams to generate personalized, conversion-focused landing pages faster than the traditional design-and-engineer approach. By merely defining their brand, audience, and message, companies can generate landing pages from keywords, GEO pages to improve the chance of being mentioned by LLMs, and ABM microsites that allow for higher personalization for groups of leads. Most recently, through integrating with Claude, companies can create and edit landing pages using natural conversation. Its API also enables integration with workflow automation tools like Zapier and n8n, enabling teams to trigger page generation as part of any existing pipeline. Altogether, a company can automatically spin up campaign-specific landing pages, working toward Flint’s mission of making page creation as fast and scalable as modern marketing demands.

Check it out: tryflint.com

The next $1B voice AI company is being built right now. Maybe by you.

Unicorner is teaming up with ElevenLabs to give founders a real head start: 33 million credits ($4k value) and 12 months free to go build it.

Voice agents. Multilingual content. Game characters that actually sound human.

Whatever you're building, ElevenLabs has the infrastructure to support it: ElevenAgents to build intelligent, expressive voice AI agents. ElevenCreative for speech, music, image & video in 70+ languages.

Apply today for the ElevenLabs Startup Grant.

Use code 11unicorner933 and say Unicorner sent you.

Flint utilizes a 4-tier subscription-based model. The Free model allows companies to explore and test Flint. It grants 160 credits a month, a one-time design system import, and website hosting with limited bandwidth. The Starter tier is priced at $8 per month paid annually and includes 1,000 credits per month, website hosting with 10 GB of bandwidth, and a custom domain or subdomain. The Pro tier is priced at around $34 per month paid annually, and includes 5,000 credits per month, an increased 50 GB bandwidth in website hosting, the ability to host on your root domain, and Slack support. The final tier is Enterprise, offering custom credits, API-driven page creation, multi-time design system syncing, a dedicated CSM, and more.

Raised $5 million seed round from Accel, with participation from Sandberg Bernthal Venture Partners and Neo
Officially launched its product this February
Increased conversion rate by 50% with Google Ads and influenced seven figures in ARR for Graphite, the AI code review platform previously covered by Unicorner
Built 341 personalized account-based pages in five minutes through a single CSV upload for Amigo, a healthcare company
Tripled paid media conversions and saved over 70 hours in manual labor for Tandem, a healthcare/prior authorization software platform, by building 14 ad-optimized landing pages in one month

Michelle Lim, co-founder and CEO of Flint, has spent much of her career building products and figuring out how they actually reach users. She started working as a software engineer at Slack and Robinhood, building products that have scaled to millions of users.

These roles prepared her for her next endeavor, co-founding and serving as the CTO at Pillnotes, a doctor-patient communication app geared toward improving patient education. She built the Pillnotes app from scratch and gained early stage startup experience, marking a step towards her path to becoming a founding CEO.

From Pillnotes, Lim joined Warp, a platform for coding with AI agents, as its second employee and first software engineer. She was heavily involved in the technical side, building the app from scratch once again and developing the graphics pipeline and block sharing, which allows users to share and reuse commands.

As Warp prepared for its private beta, Lim slowly transitioned from engineering to go-to-market work. She began by leading early launch efforts, setting up feedback channels and user interviews, and eventually became Head of Product and Growth. During this transition, she noticed that “launching even a single A/B test on [its] website was a game of telephone between five teams.”

Recognizing this gap, Lim co-founded Flint with Max Levenson, a former software engineer and tech lead at Nuro, a self-driving technology company, to build autonomous websites for marketing.

With the global digital advertising spending market size reaching $650 billion in 2025 and projected to surpass $1.48 trillion by 2034, it’s evident that companies are scouring, now more than ever, for ways to stand out and connect with consumers on a deeper level. Yet, despite the massive budgets poured into ads across Google, Meta, and more, most leads still end up on a one-size-fits-all landing page. Regardless of a customer’s need, they are met with the same general information. As a result, companies are left with an average conversion rate of just 6.6%.

Flint is targeting this niche of marketing within the emerging AI-generated web. It offers companies unique landing pages for each of their ads, keywords, and audiences, thereby providing leads with personalized advertising. In doing so, Flint claims improved conversion rates (double or triple in some cases) and, ultimately, optimizes advertising budgets.

Multiple variations of the same landing page illustrate audience-specific content.

The reason most companies only have a few landing pages to begin with is that the process is difficult and time-consuming. It requires manual processes and coordination between marketing and engineering teams. The average timeline to build a simple landing page is 1 to 3 days. With Flint, hours can be saved. Flint can generate landing pages in minutes, and in one case, saved 70 hours of manual website work.

With CMOs reportedly spending around 7.7% of company revenue on advertising on average, every small improvement matters. Since AI personalization can improve conversion rate, with Flint replacing generic landing pages with a tailored experience that directly ties to the expectation from the ad, companies can see significant increases in revenue. And by being one of the first to explore this niche in a world where almost all businesses have websites, it is well-equipped to scale greater. With strong potential in a massive market, we believe Flint can become a Unicorn.

Flint Raises $5 Million in Seed Round [The SaaS News]
Introducing Flint MCP and API [Flint Blog]
Sheryl Sandberg-backed Flint wants to use AI to autonomously build and update websites [TechCrunch]
Our Investment in Flint: Marketing Superhuman-as-a-Service [Accel]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

DeepCorner: SpaceTech

Mon, 06 Apr 2026 15:01:00 +0000

Flash back to last week. You hold your breath as the Artemis II launches, on its way to become the first space mission in over 50 years to return to the moon.

Yet, there’s so much about the world of space and the mechanics behind the industry that remains opaque.

With that, we introduce our latest issue of DeepCorner: Unicorner’s occasional deep dives into all things deep tech. Today’s focus? The world of space.

Missed the last DeepCorner? Check out some of our past deep dives:

April 29 (SF): We’re hosting our first AI Champions dinner, in partnership with Bolt.new and MiniMax. If you’re spearheading AI usage at your company, this dinner is for you. Apply to join us here.
April 30 (SF): We’re hosting a seed+ founders dinner for 20 founders with Fidelity Private Shares. Invites are extremely limited; apply here.

Ethan and Arek 🦄

The next $1B voice AI company is being built right now. Maybe by you.

Apply today for the ElevenLabs Startup Grant.

Use code 11unicorner933 and say Unicorner sent you.

^{_{This is sponsored content.}}

The last space race was won by governments. The next one will be won by founders.

For most of human history, space was the exclusive domain of nation-states with unlimited budgets and geopolitical axes to grind. NASA put humans on the Moon in 1969. The Soviet Union built the first space station in 1971. But behind every mission was a government check, timeline, and agenda.

That era is over. The privatization of space is underway. And the companies being built right now will define who controls what happens in the exosphere.

Humans went from the first flight in 1903 to landing on the Moon in 1969, one of the most compressed technological leaps in recorded history. Sixty-six years separated Kitty Hawk from the Sea of Tranquility, a timeline that makes the achievement feel almost implausible.

It happened because of the Cold War.

From the late 1950s onward, space became the highest-stakes theater in the ideological competition between the U.S. and the Soviet Union. Every launch was a proxy argument for which system of government could produce greater human achievement. With unlimited political will and the budget to match, NASA's funding peaked at an astounding 4.5% of the federal budget in the late 1960s.

Once the U.S. crossed that finish line, the urgency evaporated. Without a rival, NASA's budget entered a decline, falling below 0.5% of the federal budget by the 2000s. The Space Shuttle, conceived to make orbit routine and affordable, became a symbol of the structural difficulty in reducing the cost of access, while operating costs remained high. In the end, human spaceflight became confined largely to servicing a space station that was itself as much a diplomatic instrument as a scientific one.

Four forces have converged to make this moment different.

The most important is launch cost. Getting a payload to orbit once cost tens of thousands of dollars per kilogram (the Space Shuttle costing $54,500 per kilogram), while today, on a Falcon 9, it costs roughly $2,700 per kilogram. Starship, SpaceX’s upcoming launch platform, is targeting an order of magnitude lower than that. When access to orbit gets this cheap, previously cost-prohibitive missions and use-cases become viable and attract venture capital.

The second driver is miniaturization. The same semiconductor advances that made smartphones more powerful than 1990s supercomputers made it possible to build CubeSats and SmallSats, low-cost satellites as small as a shoebox and often launched in groups. They have allowed companies to operate constellations up to thousands of units for what a single traditional satellite once cost.

The third is AI. Machine learning has enabled autonomous mission planning, real-time onboard data processing, predictive maintenance for satellite hardware, orbit optimization, and analysis of the enormous quantities of Earth observation data now being generated. Some companies are processing geospatial data directly on the spacecraft itself, rather than downlinking hundreds of gigabytes. The ability to derive intelligence at the edge changes the cost structure.

The fourth is geopolitics. More on this later.

Altogether, the space industry expected to almost triple from $630 billion in 2023 to $1.8 trillion by 2035.

To understand where the opportunities are, imagine the space economy like any technology stack. There are layers and value accumulates differently at each.

Launch is the foundational layer. SpaceX dominates it today in a way that is difficult to overstate, and its position in reusable heavy lift is essentially unassailable in the near term.

The meaningful competition is at the small and medium lift end. Rocket Lab has established a durable position, while Stoke Space is building a fully reusable rocket with a unique staged combustion architecture. Several others are pursuing different approaches to reusability and cost reduction. All of these companies are betting that demand for dedicated launch capacity will grow faster than rideshare slots on larger vehicles can serve it.

The launch layer is now consolidating, with companies prioritizing reusability better positioned to survive.

The satellite layer is the next up, and it’s where the first generation of space unicorns emerged. Starlink changed the connectivity conversation permanently. Before it, satellite internet was slow, expensive, and mostly used by ship captains and Antarctic researchers. Today, it serves millions of subscribers and is generating real revenue.

The interesting startups are in niches Starlink cannot or will not serve. Some are building small geostationary satellites to bring connectivity to specific underserved regions with a unit economics model that does not require deploying thousands of spacecraft. Others are attempting to connect standard mobile phones directly to satellites. Still others are building the ground infrastructure that lets satellites communicate.

Earth observation is currently the clearest path to commercial revenue in the space economy. Why? Satellites produce operationally useful data for buyers across agriculture, insurance, finance, logistics, and government intelligence.

The global Earth Observation market is on track to exceed $8 billion by 2033. Raw imagery is already becoming a commodity, transferring margins to the intelligence layer. That might mean using AI to predict crop yields, track vessel movements, monitor deforestation, or identify early supply chain disruptions. Because this is a software/AI problem, it is subject to the same laws that make it fast to build, cheap to scale, and more profitable by margin than anything in the launch or manufacturing layers.

This is the earliest-stage layer but potentially the most consequential. As the number of active satellites grows, a set of supporting services becomes necessary, including refueling, inspection, debris removal, and orbital transfer. Unicorner companies like Orbit Fab and Starfish Space are betting on that long-term vision, building orbital refueling depots and satellite servicing vehicles. The bet, if accurate, brings a high reward.

Microgravity manufacturing is next. Certain manufacturing, like specific pharmaceutical compounds, fiber optic cables, and advanced alloys benefit from microgravity environments. Startups are currently launching small capsules to conduct manufacturing runs.

Further out still is the concept of orbital data centers, where physical advantages of space could make it cheaper to process particular workloads in orbit than on the ground. NVIDIA chips are already being run in orbit to evaluate viability. These businesses are early and speculative, but exciting nonetheless.

Any honest account of the space economy must grapple with the legal and sovereignty questions that remain unresolved.

The Outer Space Treaty of 1967, the foundational legal framework for space, was written when only governments launched. It prohibits national appropriation of celestial bodies but is silent regarding private companies. The U.S. Commercial Space Launch Competitiveness Act of 2015 established that American citizens can own resources they extract from space, but this is not international settled.

Orbital slots and radio frequency spectrum are finite, and the operators who move first claim priority. That is a meaningful part of why large constellation operators have deployed so aggressively.

And there is, of course, the Moon. The U.S.’s Artemis program has committed to returning humans to the Moon as early as 2029. The Chinese Lunar Exploration Program is operating on a similar timeline. Competition is over the lunar south pole, where water ice deposits could be converted into rocket propellant, giving the owner control of a refueling depot for deeper exploration.

The clearest opportunity is with the intelligence layer. Satellites are generating more data than the industry currently knows how to process or monetize. Winners here will build the tools to analyze and distribute insights from this data. Like other software companies, the upside to success at scale is enormous.

The second is in-space services. As of early 2026, there are over 2,500 active companies in the space tech sector. As satellite count grows, so do the markets for preventing congestion and keeping satellites operating. The companies building that capacity now are early to a market that only gets larger as the layers below them scale.

The third opportunity, further out, is in-space manufacturing and resource utilization. The economics of producing materials in microgravity or extracting resources in space may prove to be commercially significant as other innovation (such as robotics) converges.

Investors are signaling where they believe the value is concentrating through their funding decisions. Late-stage funding as a percentage of total space investment has reached its highest level in a decade, which means capital is moving away from early bets on unproven infrastructure and toward companies that have demonstrated commercial traction. Space companies raised over $12 billion last year, with 2026 off to a strong start.

The pattern here mirrors how value migrated in the internet era. Building the fiber and the servers was necessary but the returns went to the companies that built what ran on top. Think Amazon, Google, Meta, and more. The same dynamic is playing out in orbit.

The funding numbers and market projections are compelling. However, the space economy comes with a set of structural risks that do not have clean analogies in software or even most other DeepTech sectors.

Space is expensive in ways that compound. Hardware must be designed to survive launch forces, radiation exposure, extreme temperature swings, and, depending on the mission, years of operation with no possibility of physical maintenance. The result is engineering timelines and cost structures that are fundamentally different from software-style venture returns. Moreover, a failed satellite represents total destruction of the asset, with no patches or rollbacks.

Many space startups are burning significant capital for years before generating any revenue, and occasions like the 2022 to 2023 VC pullback hit several high-profile space companies hard.

In just the U.S., regulations are patchwork, with launch licensing running through the Federal Aviation Administration and radio frequency coordination running through the Federal Communications Commission and the International Telecommunication Union. Beyond the U.S., countries have different national space laws, and the international legal framework for resource extraction and orbital governance remains contested. For a startup trying to move fast, regulatory uncertainty is a strategic risk.

The technical expertise required to build space hardware is scarce. Most of it lives inside legacy aerospace primes like Lockheed Martin and Airbus, government labs, or a handful of elite universities, and startup recruiting requires competitive compensation as well as a convincing mission and a credible path to a business.

The AI and software talent that is central to the intelligence and data layers of the space economy is more available, but it means direct competition for talent with companies that offer lower risk and greater compensation.

Every major technological revolution has required physical infrastructure before it could produce software-layer returns. Think fiber for the internet or data centers for cloud computing. This time, the foundation sits 250 miles above the Earth's surface.

With demand already embedded in the global economy, today’s founders are establishing the operating system for the next century of human civilization.

From bankrupt to $20 million ARR in 2 months

Thu, 02 Apr 2026 15:01:00 +0000

She was saying something. Eric Simons wasn't listening.

Sitting in the back of the classroom, he tapped his pen against the desk, eyes drifting to the window. His absent mind caught fragments, something about chemical reactions, before sliding away again. He glanced at the clock: 3 minutes, and he'd be free. After what felt like an eternity, the bell rang. The young, blond student with spiked hair rushed to gather his things and head home.

"Eric, do you have a minute?" asked the chemistry teacher.

She's going to call me out for doing nothing in class. Again. He sighed quietly, looking the other way so she couldn't see his face. "Sure." He walked over to her desk.

The question she asked would be the beginning of Eric's career as a serial entrepreneur and, some years later, would lead to the company that went from $0 to $20 million in ARR in just two months.

Today, we’re excited to unveil Unicorner’s second publication, Zero to Unicorn.

Every other Thursday, we’ll send you a deep dive sharing the life stories of billion-dollar founders, investors, and operators.

We’ll tell the untold tales of the miraculous founders who have made it to the pinnacle of their industries against all odds. That includes the impossible pivots that arose out of necessity, the years where the product was great but the business wasn't, and the moments our protagonist almost walked away. And expect the highly coveted, hard-won lessons they learned along the way.

We’ll also host regular, in-person Zero to Unicorn founder fireside chats in SF and NYC, similar to our past conversations with founders like Tyler from beehiiv and Eric from Bolt.

That’s all for now. We hope you enjoy. Back to Eric’s story!

Ethan and Arek 🦄

At the time of that fateful day, learning to program was a challenge. One Google search in and you'd be inundated with words like HTML, frontend, backend, databases, React, and JavaScript. Even the smallest idea felt like climbing a mountain. To no one’s surprise, most of them didn’t get built.

Today, instead of trying to build the app brick by brick, you simply navigate to bolt.new, describe what you want, and a full project appears in front of you.

When Bolt was launched in October 2024 with nothing more than a single tweet, the internet lost its mind. Over 60,000 people signed up on day one.

Within two months, the company went from $0 to $20 million in ARR, a growth curve so steep that even seasoned investors said they'd never seen anything like it. Today, more than 10 million people have signed up to use Bolt, including employees at 75% of Fortune 500 companies.

Hidden behind this seemingly overnight success story is years of hard work and tough times. How did they do it? In our inaugural article of Zero to Unicorn, we tell the story behind how Bolt went from a tweet to tens of millions of users while trailblazing a new industry and democratizing building for all.

Young Eric Simons and Albert Pai. (Courtesy of Bolt)

Long before Bolt, our story begins in a suburb of Chicago with two little boys who liked computers a little too much.

Eric Simons and Albert Pai were drawn together because of a mutual love of computers at age 13. Soon, they’d taught themselves to code to create games, and the bond stayed for life.

Not long after, they had already shipped their first product, a cloud storage service for students to sync their schoolwork between home and school. It was, essentially, Dropbox-meets-Pinterest years before either was developed. It spread across school districts in Illinois and the Midwest.

Although he created a great product that helped a lot of schools, Eric was not an A+ student. As he describes himself: "I couldn't stay focused in school, I wasn't interested in homework, and I wasn't motivated by grades."

In 2010, during his junior year in high school, Eric was asked a question that would change his life.

"What would make you interested in learning what I’m teaching?"

It was his (desperate) chemistry teacher. Naturally, being a computer lover, Eric’s first suggestion was "Let’s get everyone working together on computers – I'll even build the software for us to use."

This was the humble origin of ClassConnect, officially launched in 2011, which he spent the next two years of high school building, eventually making him a kind of celebrity among startup entrepreneurs in Silicon Valley.

The idea of ClassConnect was to allow students to work together using videos and websites, making them learn better.

Ironically, despite not being a straight-A student, Eric developed a passion for education.

At 18 years old, with high school ending, the duo was facing a challenge common for young adults. College was too expensive, but rather than bite the bullet, they took a different approach befitting two ambitious teenagers with a knack for technology and no money. They applied to accelerators.

Thanks to ClassConnect, they were accepted into Imagine K12, a startup accelerator focused on startups building products for primary and secondary education (K-12), which awarded them $20,000 and a reason to pack their bags for Silicon Valley around 2012. The goal was to work full-time on the solution, expanding it from a student tool into a platform that also helped teachers create, share, and discover lesson plans.

They burned through the $20,000 before the product was ready. When the money ran out and the program ended, Eric, now 19 and essentially homeless, found a creative solution to keep working on ClassConnect: squatting at AOL headquarters.

See, everyone involved in Imagine K12 had access to AOL's Palo Alto campus, since the accelerator was located there. It turns out, the badges still worked even after the program had ended. The building had couches, a gym, and showers…

At 19, that was good enough housing.

For four months, he coded from 9 a.m. to midnight and slept on AOL's furniture until someone finally noticed and threw him out.

The story could have ended there with a broke 19-year-old with a half-built product and no place to sleep. But CNET caught wind of it and published an article, and the internet loved it. Even CNN picked it up, and Eric found himself famous as “the AOL squatter,” becoming the protagonist of one of those stories the startup world can't quite let go of. To this day, it follows him into talks and interviews, a reminder that in Silicon Valley, sleeping on a couch in a building you're not supposed to be in can, under the right circumstances, make you a legend.

That notoriety didn't hurt, and Eric was determined. His hard work with ClassConnect during this period paid off with $50,000 in seed funding.

Unfortunately, this is not our successful story yet. ClassConnect, later called Claco, didn’t ramp up. With funding running out, a mix of more complex competitor platforms started to appear, and the difficulty of scaling in EdTech, with its rigid cycles, led the company to wind down around 2012. It was an early lesson: building something valuable doesn’t guarantee people will use it. Distribution and ease of adoption matter just as much. It’s a lesson that would come back later, in a very different context.

Money was tight, but the ambition to keep creating was still very much there.

As Eric recalls, this was a time when the duo was constantly testing new startup ideas and building web applications for them. They soon realized they had become experts at web development and had the insight to create an online platform teaching what they knew, hoping it would generate enough income for them to keep trying ideas.

This idea became Thinkster, launched in 2014, an online platform designed to teach web development. And it worked, eventually reaching over 100,000 learners.

Thinkster was acquired by Joe Eames in 2019, a developer educator Eric and Albert trusted enough to hand the keys to. In a Medium post sharing the news, Eric signed off with, "Every new beginning comes from some other beginning's end," which, for the team’s story up until this point, felt about right.

For the duo, it was less an ending than a redirect because, with Thinkster taken care of, they could turn their full attention to their next company. That’s because, while working on Thinkster, there was one recurring problem they encountered: learners needed to set up a local environment on their computers to start programming, but doing so was challenging, and most beginners quit before ever writing a single line of code.

They wondered… what if the entire development environment could run inside the browser, so new learners wouldn't have to install anything? That would eliminate the biggest challenge standing between their students and their lessons.

Someone must have done this already, right?

They looked around.

No, no one had.

Thus begins the story of StackBlitz, the duo’s focus over the next half a decade that unlocked the insight and competitive advantage behind Bolt.

StackBlitz was born to make coding more accessible by creating a full-stack web development environment in the browser, just as Figma and Canva brought design to the web.

It was founded in 2017, before Thinkster’s acquisition. At this point, the team was building StackBlitz alongside Thinkster, addressing the recurring pain point through StackBlitz.

To make it possible, they built a technology called WebContainers, a browser-based runtime that lets you run full Node.js development environments (the runtime that learners need to install to begin programming) directly inside a web app. It took years of trial and error to make it work (remember, this technology simply didn’t exist back then!), but they eventually got it up and running. In the end, they had their flagship product, a web-based IDE for aspiring and working developers.

Developers were supposed to love it, in theory. But in practice, getting them to actually leave their local environments was nearly impossible.

If you know one, you’ll understand. Developers are creatures of habit. They have setups they've spent years perfecting. A browser-based alternative was a cool novelty, sure, but they weren’t about to start from scratch there, no, thank you.

The cloud IDE category, as Eric would later describe in a Forbes Talks episode, turned out to be "a mirage of a market."

The founders ran into the same problem twice, in very different ways.

With ClassConnect, teachers and students liked the product, but adoption depended on institutions, slow-moving systems that made it hard to scale or generate revenue. StackBlitz hit a different wall, with developers liking it too, but not willing to change the environments they had spent years refining.

In both cases, the product worked and people liked it. However, it failed at different layers of growth.

The real challenge wasn’t building a good product for users; they had already done that twice. It was figuring out how to make this great product easy to distribute and adopt, without friction slowing it down.

Seven years in, StackBlitz was in decline, doing around just $500,000 in annual revenue. The technology was really amazing. The business was not.

So, in 2023, the team found themselves facing the hardest choice of their lives.

The team, now in 2024, was faced with a moment of truth. With a year of runway, either they would find a way to make StackBlitz profitable, or they would start dissolving the company.

2024 was also the year Eric became a father. With both a baby and a company necessitating his attention, Eric found comfort in pursuing yet another challenge. He committed to training for a full Ironman triathlon despite having never run a marathon before.

It sounds irrational, but it wasn’t. He had made peace with the idea that the outcome might not be in his control. Persistence, he understood, is only as good as what's driving it, and what was driving him wasn't success, but a genuine desire to see what he had built come to life. If it didn't work out, he could live with that. What he needed to focus on was how hard he worked at everything, including his health, his family, and his work, before the clock ran out.

"I've never been happier in my life… Really investing in myself, my co-workers, my products, my family, my business, with just such intense dedication. I was okay with whatever the outcome was."
Eric Simons

Despite his mindset, the team still needed a miracle.

With nothing left to lose, for all of 2024, they experimented. A lot.

Bolt was the last experiment on the list before the end of the year, and with it, StackBlitz.

When we set out to create Zero to Unicorn, we were ambitious. We wanted to produce a publication that was never made before.

Just like our IRL Zero to Unicorn firesides, our mission was to unearth the untold founder stories, all of the true ups and downs, the mistakes as well as the successes, and share the hard-earned lessons that were learned along the way, of the very best.

Turns out, we had too much for just one article.

In Part II of Bolt’s story, we’ll cover the miraculous turn around that not just saved the company, but helped usher in a whole new paradigm of engineering. Plus, we’ll tell all of the untold lessons from years of entrepreneurship that bolstered Bolt’s success.

Stay tuned!

Radical AI

Mon, 30 Mar 2026 15:00:00 +0000

Happy Monday.

Missed the news on our socials? We just officially announced our second newsletter: Zero to Unicorn.

For the past five years, we’ve covered the stories of promising startups. After a number of oversubscribed IRL events with billion-dollar founders (like Bilt Rewards, Kalshi, and beehiiv), we’ll now also be diving into the journeys of the unicorn startups that have made it.

First edition drops in your inbox on Thursday. Keep your eyes peeled, and your emails refreshed.

Arek and Ethan 🦄

From replacing stone with bronze to turning raw sand into silicon chips, materials science has always been the invisible backbone of humanity’s greatest technological leaps. Yet, for all its transformative power, the process of developing these innovations remains remarkably inefficient. The typical path to developing a novel material requires scientists to first understand the underlying chemistry and physics, then determine the process of manufacturing them at scale, and finally put them through rigorous stress testing. That process routinely lasts up to 20 years, burning through extensive R&D funding.

Radical AI is rethinking the scientific method by building a closed-loop system for material discovery and development, where AI designs materials, tests them, and learns from the results in a continuous feedback cycle. Specifically, it uses a process called inverse design, where it starts from a target property, such as thermal resistance from a hypersonic vehicle. Then, using machine learning, it works backwards to reduce the enormous compositional search space to the most promising candidate alloys. Finally, its largely autonomous, self-driving lab synthesizes and validates those materials. The resulting experimental data is captured and fed directly back into its models so the AI scientist can learn in real time from each experiment, continuously refining its intuition and accelerating the next cycle of discovery.

Radical’s goal is not only to make discoveries in materials science but also to build breakthrough materials that could revolutionize industries, such as novel alloys for hypersonic aircraft, reusable space systems, and next‑generation nuclear fusion reactors.

Check it out: radical-ai.com

The next $1B voice AI company is being built right now. Maybe by you.

Apply today for the ElevenLabs Startup Grant.

Use code 11unicorner933 and say Unicorner sent you.

Radical is a full-stack materials company discovering high value, application-specific materials and monetizing them by either selling proprietary materials/components or selectively licensing the underlying technology to partners. Its primary customers include defense contractors, semiconductor fabs, hypersonic flight companies, and nuclear fusion companies that demand niche materials.

Raised a $55 million seed+ round led by RTX Ventures, alongside NVentures (NVIDIA's VC arm), noa, Eni Next, Infinite Capital, and AlleyCorp
Actively testing tens of thousands of materials per week, with programs in hypersonics and nuclear fusion materials

As a materials science PhD candidate at Rice University and researcher at the US Army Research Lab, Radical’s CEO Joseph Krause experienced firsthand how slow, fragmented, and manual materials R&D is. As an investor at AlleyCorp in 2022, he watched as AI continued to transform industries while material science remained largely untouched, despite being the critical bottleneck for frontier technologies.

While brainstorming with co-founder Jorge Colindres, fellow investor at AlleyCorp, the two concluded that the real opportunity lay in automating the discovery and commercialization of new materials with a fully integrated, AI-native approach. According to Krause, self-driving labs were already proving their value in biopharma, and it was clear that materials science needed the same. Knowing this, Krause and Colindres flew to California to meet the person who would become their future co-founder, Gerbrand Ceder. Cedar leads the Computational and Experimental Design of Emerging Materials Research Group at UC Berkeley and has been actively building an AI-driven materials science lab. The three quickly realized they shared the same conviction that the shift was inevitable and decided it was time to build Radical.

Radical AI's customers span four high-value verticals, each with significant materials demand. The global semiconductor materials market was valued at roughly $72 billion in 2025 and is projected to reach about $104 billion by 2034, with the compound semiconductor sub-segment estimated at $67.69 billion. The aerospace and defense materials market is expected to reach $49 billion by 2035, and the broader hypersonic technology market, where thermal protection materials remain a critical bottleneck, was valued at $6.7 billion in 2024 and is projected to reach $12.4 billion by 2033. Nuclear fusion materials represent a smaller but fast-growing niche driven by the need for plasma-facing components that can survive extreme conditions. This makes up a $165 billion opportunity altogether.

The AI materials discovery field is evolving quickly, and competitors now fall into two clear camps: software platforms that use AI to accelerate R&D but do not operate physical labs, and automated labs that pair machine learning with robotics. Among its competitors, Radical's edge comes from its operational autonomous lab, which generates proprietary experimental data that compounds its models with every run. This closed-loop system, where each experiment refines the next prediction, is designed to be difficult to replicate and bound to grow more defensible over time. As Krause explains, they are testing tens of thousands of materials per week and are developing proprietary materials in hypersonics and nuclear fusion.

The autonomous lab where robotic synthesis meets a closed-loop AI system, each experiment accelerating the next

A core challenge across the sector to consider is the speed of adoption. While the demand undoubtedly exists, commercializing materials remains slow and resource-intensive. For Radical, this hurdle is especially notable since its defense, semiconductor, and fusion clients impose long qualification cycles before contracts convert to revenue. For instance, even if AI accelerates the R&D of these materials, actual customer adoption still requires integration into existing product architectures, validation of performance at the system level, and passing formal qualification (the process of certifying the material for real-world use), which collectively stretch into several years. Assuming the materials are validated and can effectively serve their purposes, the broader question then becomes whether AI-driven discovery can close the gap between insight and deployment fast enough to justify the capital now flowing into the field. Whoever can meet that demand will be a runaway winner, and Radical is positioning itself to be one of them.

Material Abundance: Radical AI’s Closed-Loop Lab Automates Scientific Discovery [The Cognitive Revolution Podcast]
Everyone’s Misunderstanding AI’s True Potential | Radical AI, Joseph F. Krause [EO]
Tech Demo: The Age of Autonomous Materials Innovation [Special Competitive Studies Project]
DeepCorner: Advanced Materials [Unicorner]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Tether

Mon, 23 Mar 2026 15:00:00 +0000

Noticing something a little different about today’s article?

We’re excited to announce that Unicorner’s flagship Startup of the Week newsletter is rebranding to The Highlight, with the mission of highlighting one hot startup every Monday morning.

Nothing is changing programming-wise, but we have another pretty major newsletter announcement coming very soon… stay tuned.

Ethan and Arek 🦄

Tether addresses one of the most fundamental challenges in the clean energy transition: grid balancing. Because renewable energy sources like solar and wind are intermittent by nature, grid operators must constantly match supply with demand in real time, historically relying on fossil fuel peaker plants to fill the gap. Tether’s solution is a prediction and optimization engine for EV charging infrastructure. Its platform uses machine learning to predict periods of grid overutilization or underutilization ahead of time, then dynamically adjusts EV charging speeds across its partners' networks to absorb excess supply or reduce demand accordingly. That aggregated flexibility is bid on European balancing markets (where grid supply and demand are adjusted in real time), generating a new revenue stream shared back with charge point operators, vehicle manufacturers, and fleet managers.

The platform integrates with existing charging infrastructure, requiring no hardware changes. Vehicle-to-grid capability, where energy flows bidirectionally between cars and the grid, remains a longer-term ambition as compatible EVs become more widespread over the next five to ten years.

Check it out: tetherev.io

The next $1B voice AI company is being built right now. Maybe by you.

Apply today for the ElevenLabs Startup Grant.

Use code 11unicorner933 and say Unicorner sent you.

Tether operates on a revenue share model, taking a percentage of the revenue generated by bidding aggregated EV charging flexibility on energy markets. The platform also generates recurring fees through data storage and analytics services.

Raised $1.5 million to date, backed by Draper B1, KFund, Enzo Ventures, Axel Carbon Capital, Inclimo Climate Tech Fund, and Earth Capital
Signed proof of concept agreements with two large European vehicle manufacturers and has multiple charge point operator clients in Sweden
Took part in the spring 2023 Techstars Audi Founder Innovation Lab class
Selected for MobilityXlab, working alongside Volvo Group, Polestar, Zeekr, and Ericsson
Named a finalist in the 2025 European Startup Prize for Mobility

Tether does not have open positions right now, but the company is constantly looking for new partners, engineers, and developers who align with its mission. If you are interested, reach out to us by replying directly to this email!

The throughline of Luis's career is the power grid. Even while studying electrical engineering at Florida International University, where he graduated cum laude, he was already focused on how energy systems operate and interact. That interest deepened during his dual master's in Renewable Energy through EIT InnoEnergy, where he studied Energy Innovation at KTH Royal Institute of Technology in Stockholm and Energy Engineering at the UPC Universitat Politècnica de Catalunya in Barcelona, alongside an entrepreneurship certification at Esade Business School.

He started at Florida Power & Light before moving to GE, where he worked across both the Power and Solar divisions, building expertise in energy controls, how power plants interact with one another, and the fundamental mechanics of grid stability. He then joined Northvolt in Stockholm as an Integrated Systems Engineer, working on battery energy storage systems. It was during his master's thesis that Rivas first studied using EVs as balancing assets for the power grid, the insight that would eventually become Tether.

Rivas founded Tether in 2022, going through the Esade eWorks accelerator before being selected for the 2023 Techstars Audi Founder Innovation Lab. He is also a recognized Future Energy Leader by the World Energy Council.

The clean energy transition has an underappreciated infrastructure problem. Renewables now make up 43.2% of the European power mix, but their intermittency creates constant volatility on the grid. Every time cloud cover hits a solar farm or wind drops, grid operators need to compensate in real time, historically by spinning up fossil fuel peaker plants that exist solely to absorb those swings. Demand for flexibility in Europe is expected to more than double to keep pace with an increasingly intermittent power system, and the conventional answer, utility-scale battery storage, is expensive to build and slow to deploy.

What Tether recognized is that the storage already exists. There are millions of EVs plugged in across Europe at any given moment. In 2025 alone, over 2 million battery electric vehicles were registered, reflecting the rapid scale of adoption. Each one is a controllable load that can be sped up or slowed down in response to grid conditions. In aggregate, they represent a balancing asset of real scale that does not require new infrastructure to unlock.

Tether's platform sits between the charge point operator and the balancing market, predicting grid stress ahead of time and dynamically adjusting charging speeds across its network to absorb or reduce demand accordingly. The revenue generated from bidding that flexibility on European balancing markets is shared back with operators, creating a new income stream on hardware they already own. The software-only model employed by Tether, which has no hardware requirement and an integration with the charger operators that takes under a week, is the right go-to-market for a fragmented operator market where capital intensity and thin margins are chronic pain points. European grid operators could benefit from a projected €4 billion in annual savings from optimizing the grid and making use of flexibility from available assets, and Tether is positioning itself as the software layer that captures a share of that value.

Tether’s platform turns EVs into grid-responsive energy assets.

By 2030, EVs on European roads could contribute up to 4% of Europe's annual power supply, enough to power 30 million homes, and by 2040, if all EVs are capable of bidirectional charging, over 10% of Europe's power needs could be stored and reinjected when needed. Tether's current smart charging model does not require V2G-capable vehicles, but the roadmap toward bidirectional capability as compatible EVs become widespread over the next five to ten years substantially expands the revenue opportunity per vehicle and per market. The EU has signaled that fully interoperable V2G solutions are expected to be widely available by 2027, giving Tether a clear regulatory tailwind for that transition.

Europe's fragmented approach to electricity supply security cost the continent almost €11 billion in 2024, and navigating each country's balancing mechanism adds complexity to every new market Tether enters. The charge point operator market is capital intensive with thin margins and commoditized hardware, which constrains how much flexibility value Tether can extract per operator. EV adoption, while structurally growing, has shown signs of slowing in key markets, directly affecting the density of the fleet Tether can aggregate. That said, the core business does not require V2G or rapid EV growth to function today. The immediate opportunity in smart charging coordination is real and growing, and the V2G roadmap is upside.

Spanish clean energy startup Tether secures €1.3 million to turn EVs into the “world’s largest connected battery” [EU-Startups]
Everything You Need to Know About V2G [Virta.Global]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Daytona

Mon, 16 Mar 2026 15:00:00 +0000

25 investors. Amazing partners (shoutout Mercury and Sydecar). Incredible food.

In other words: it’s fair to say Unicorner hosted a great SXSW omakase last night.

Among other conversations: we think we’re certified tokenheads, and so are many other VCs at this point.

Tokenhead • noun
To be addicted to the usage of Claude, ChatGPT, and similar LLMs to the extent of expending a vast multitude of tokens.

We’re in Austin until Thursday night. Want to grab a coffee while we’re here? Reply to this email, and we’ll make it happen. 🤝

Arek and Ethan 🦄

Daytona is an open-source platform for running AI-generated code inside isolated sandbox environments, secure mini-computers that let AI agents execute code without affecting other systems. It treats each sandbox as a composable computer, meaning an agent can create, configure, save, copy, and resume an environment through an API as it completes a multi-step task. It is built for AI agent workflows that need access to information and long-running execution within a persistent environment. The platform supports snapshotting, recovery, and forking, so teams can save and restore progress or duplicate environments for experimentation. This helps reduce wasted compute when agent tasks require multiple attempts, and it allows teams to run many sandboxes concurrently for large-scale workloads.

Check it out: daytona.io

Daytona uses usage-based pricing. It includes $200 in free compute and 5 GB of free storage, then bills per second for compute and memory, with metered storage beyond the free tier. It also offers enterprise options for larger deployments, including on-premises setups and a startup program with additional credits.

Raised $24 million in Series A funding in February 2026, led by FirstMark Capital
Raised $5 million in seed funding in June 2024, led by Upfront Ventures, and $2 million in pre-seed funding in November 2023
After rebuilding and relaunching for AI agents in late April 2025, the platform crossed $1 million in ARR within two months, demonstrating a rare pace of adoption among infrastructure products
Case study metrics include 4,000 sandboxes created per month, more than 660 hours of sandbox runtime per month, 10 to 30-minute sessions supported, and under 100 milliseconds of sandbox creation time, according to LangChain, an open-source framework for building LLM agents
More than 65,200 GitHub stars on the main open-source repo as of March 15, 2026

Open roles:

Daytona is also offering a startup program offering $10,000 upfront and up to $50,000 in Daytona credits.

Daytona was founded by CEO Ivan Burazin, CTO Vedran Jukic, and Chief Architect Goran Draganić. Burazin has worked in developer tools since 2009. His background includes co-founding Codeanywhere, a cloud-based development platform that provides browser-based integrated development environments, and building Shift, a developer conference founded in Croatia that grew into one of the largest developer events in Southeastern Europe and was later acquired by Infobip.

Daytona started in 2023 as an enterprise development environment for human engineers, aimed at companies that wanted a secure, self-hosted environment for management. Inbound demand from teams building agents exposed a gap where an enterprise dev environment could look like a fit, then fail in production, because agents need fully programmatic control, persistent state, and the ability to recover, fork, and continue autonomously. With AI agents hitting the limits of infrastructure designed for humans and traditional production workloads, the team hit an inflection point.

According to Burazin, the key insight was realizing that agents should be treated as a different type of user, with different requirements than human engineers:

That drove a high-conviction reset. The team sunsetted the earlier product, let go of the existing customer base, rebuilt from scratch in early 2025, and launched the new Daytona in late April 2025 as a runtime infrastructure built specifically for agents.

Daytona is going after a bottleneck that becomes obvious the moment agents leave the demo stage, when code execution stops being a small feature and starts looking like infrastructure. This paradigm is fraught with problems. Agents work iteratively, which requires execution that can run code and preserve context across the workflow. Stateless execution resets on every retry, which forces repeated setup, increases cost, and makes long-running workflows fragile.

What Daytona is really promising is not just safety. It is a way to give agents a real computer that they can start on demand and control through an API, with sessions that persist long enough to finish real work.

This market also has the kind of usage curve that can explode. Once a team relies on sandboxes as its execution layer, usage tends to expand with more agents, more tasks, longer sessions, and higher concurrency. That fits Daytona’s pricing model and the way these workloads scale, from computer use systems to reinforcement learning jobs that can require thousands of environments running in parallel.

Scientific workloads are an emerging example. We have genome work and drug discovery, and chemistry and pharma teams running agents inside sandboxes sized to hundreds of CPUs.

The category will be crowded because agent execution is a universal need. McKinsey’s 2025 survey found 23% of respondents say their organizations are already scaling an agentic AI system somewhere in the enterprise, and another 39% report experimenting with one. This pressure is strongest for enterprise teams running agents that execute code in production, often with sensitive data.

Daytona’s architecture shows how interface, control, and compute layers work together to create and manage secure sandbox environments for agents.

As adoption grows, more teams will try to solve execution in different ways, building on containers and serverless, buying a sandbox layer, or defaulting to whatever cloud providers and agent-infrastructure companies ship. These approaches either create a major security and reliability burden or rely on stateless serverless execution that resets environments and breaks multi-step workflows. The hard part is that execution at scale demands reliability and security controls that hold up under high concurrency. That is where Daytona stands out, because it is designed around the realities of production agent execution, not just the demo version of it.

The conviction behind that approach is simple: every knowledge worker needs a computer, and agents will need millions. Building the infrastructure that powers those computers safely and reliably is a platform-sized opportunity. Owning that execution layer puts Daytona underneath an expanding share of agent work as adoption grows.

Daytona wants to be an enterprise-grade GitHub Codespaces [TechCrunch]
Daytona Raises $24M Series A to Give Every Agent a Computer [Daytona]
How LangChain Found a Trusted Partner for Their Sandbox Needs [Daytona]
Daytona Goes Open Source: Embarking on a Bold New Journey [Daytona]
Daytona’s source code [GitHub]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Humand

Mon, 09 Mar 2026 15:00:00 +0000

The major players in HR tooling are building for office workers; Humand is building for everyone else.

Keeping you updated: we’re in Austin from March 14-19 for SXSW. Events we’re hosting (and RSVPs) are below. If you want to grab a coffee, respond to this email, and we’ll make it happen.

Ethan and Arek🦄

Sunday, March 15 (Austin): We’re hosting an emerging managers + founding GPs dinner for 25 VCs at SXSW with Mercury and Sydecar. RSVP here.

Humand is an all-in-one platform that combines internal communication, HR management, operations, talent development, and company culture tools.

Unlike many HR tools that are built primarily for office environments, Humand is designed for deskless workers, such as teams working in factories, retail stores, hospitals, construction sites, and field operations. Users do not need a corporate email address to access the system, bypassing one of the main points of friction with HR tools.

The product is organized into 30 modular features, each designed to handle a specific workplace task. Organizations can manage activities such as company announcements, internal chat, digital social feeds, onboarding workflows, training programs, performance reviews, time-off requests, employee records, and more. Together, they cover the full employee lifecycle, from recruitment and onboarding through daily operations and employee engagement.

Humand also includes an AI assistant called Sammy. It answers internal questions about company policies and processes, automating routine requests that would otherwise require HR intervention.

Check it out: humand.co

Humand operates as a B2B SaaS platform. The product is available as a modular structure that allows organizations to start small and add more modules as their needs evolve.

Announced a $66 million Series A on Feb. 23, 2026, co-led by Kaszek and Goodwater Capital, with participation from Y Combinator and technology founders such as Dropbox co-founder Arash Ferdowsi, Lyft co-founder Rajat Suri, and Mercado Libre founder Marcos Galperin
Took part in the Y Combinator W22 batch
Signed a deal with OXXO, one of the largest companies in Latin America with over 140,000 employees, in September 2023, in addition to signing deals with other companies
River Plate, one of Argentina’s most successful soccer clubs, joined as a client in 2024
Used by 1,500+ companies and 1.6 million employees around the world, with clients including Siemens, Domino’s, and The Home Depot

Humand founders Nicolas Benenzon and Geronimo Maspero met while studying at the Buenos Aires Institute of Technology. Their interest in building technology companies grew after attending a talk by Marcos Galperin, founder of Mercado Libre, the largest e-commerce and fintech platform in Latin America, a combination of Amazon and PayPal.

In 2018, the team participated in JPMorgan Chase’s Code for Good, a global hackathon where developers build technology solutions for nonprofits. Their project, an internal social network designed to help NGOs connect with job opportunities and talent, won the competition.

The app quickly started drawing attention to the duo behind it. In 2019, the Chief HR Officer of ArcelorMittal Acindar, a major steel producer in Argentina, invited the team to propose and build a solution for improving internal communication at one of its plants.

While working on this project, Benenzon and Maspero noticed a recurring problem: companies with large frontline workforces (such as warehouse staff and factory operators) struggled to communicate digitally with their employees working outside office environments.

For these deskless workers, paper processes, notice boards, and outdated systems that hadn’t been updated in years were still the primary way of communicating.

ArcelorMittal Acindar gave the team 20 days to build a working prototype. The first version launched as a pilot with 200 employees. In 2020, the COVID-19 pandemic accelerated the adoption of the platform, and it became the primary communication channel for more than 3,000 employees.

The team applied to Y Combinator twice, once in 2020 and again in 2021, but both applications were rejected. The feedback they received was to return once they had more customers and a clearer, scalable model.

In Winter 2022, Humand was accepted into Y Combinator and raised its first funding round of $2.5 million, backed in part by the founders of companies like Dropbox and Lyft. Since then, Humand has expanded to large enterprise clients across multiple industries, including organizations with workforces exceeding 100,000 employees like Siemens, one of the world’s largest German industrial technology companies.

Roy Anderson is an employee at the Scranton branch of Dunder Mifflin in The Office. He works on the warehouse floor under Darryl Philbin.

Unlike Jim Halpert, Dwight Schrute, and Stanley Hudson, Roy’s work doesn’t revolve around emails and meetings. He spends his time on the warehouse floor, where he loads trucks and fulfills shipments. It’s easy to imagine obstacles in his daily routine.

He arrives at work but doesn’t have a roy@dundermifflin.com email address. Updates from the company are posted on bulletin boards or shared through face-to-face interactions. There’s perhaps a computer, but it’s running an old system that crashes frequently. Time off and many other requests are filled out on paper. Training and other updates usually come as cascaded information from his boss, Darryl.

For many workers, this TV series dynamic is not an exaggeration. It actually reflects reality.

An estimated 80% of the global workforce is deskless, working in factories, retail stores, hospitals, construction sites, and logistics operations. These deskless workers are responsible for much of the daily operations of companies, but they have had limited access to digital systems.

44% of deskless workers say they expect better tools and technology to make their work easier, according to HFS Research. The same research indicates that 61% of organizations are moving toward unified digital workplace platforms. Together, these numbers show why companies are increasingly looking for solutions that can better connect and support all their employees.

This change comes with, of course, several challenges. One of them is that much of enterprise software was designed with a specific type of employee in mind: the one sitting behind a desk with a computer. Think Slack, ClickUp, Microsoft Teams, Zoom, and many more.

Humand positions itself around this gap. The all-in-one platform is designed as a mobile-first workplace system that connects frontline workers directly to company communication and HR processes.

The platform itself is comprehensive, offering modules that focus on a number of key administrative functions.

An example of Humand’s onboarding workflow, where employees receive tasks and reminders directly in the app.

For example, the internal communication module includes live streaming, internal chat, digital magazines, knowledge libraries, benefits announcements, and even an internal social network for company updates. This, and all other modules, help organizations keep frontline employees informed, avoiding fragmented and disconnected information.

One other function is the company’s AI assistant, Sammy. For employees, Sammy can retrieve information or complete simple tasks quickly. For companies, it reduces support tickets and repetitive administrative work. Humand reports that the assistant can save up to two hours per employee per day by automating these common interactions.

By consolidating these functions into a single easy-to-use application, Humand aims to replace the combination of disconnected spreadsheets and outdated intranets that many companies still rely on.

The market opportunity is significant. There are 2.7 billion people globally operating outside traditional office environments, and, as mentioned, many organizations are still transitioning away from paper-based processes and fragmented software systems. Humand demonstrates traction, with millions of workers using the platform across thousands of companies in more than 40 industries, and offices in San Francisco, New York, London, and 11 cities worldwide.

Looking ahead, Humand plans to expand further into the U.S. market while continuing to grow across the more than 50 countries where its customers operate.

As CTO and co-founder Geronimo Maspero puts it, the vision is straightforward:

“…to become the default operating system for companies. We work tirelessly every day to give deskless workers the same digital access, voice, and opportunity as anyone behind a desk.”
Geronimo Maspero, Humand CTO

This includes, of course, the warehouse floor.

Humand: $66 Million Series A Raised For AI Workforce Platform [Pulse 2.0]
Humand: All-in-one app for internal communication and HR [Y Combinator]
We’ve just raised $66M in Series A | Humand [YouTube]
We’ve just raised $66M in Series A to accelerate global growth, product innovation, and team expansion [Humand]
HR app for 'deskless' workers raises $66 million [The Daily Star]
Humand Raises $66 Million To Expand AI Operating System For Deskless Workers [VentureBurn]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Resolve AI

Mon, 02 Mar 2026 16:00:00 +0000

Happy Monday.

We had a great time in SF last week. A room full of founders and even better conversations? Count us in, always.

We’ll be back soon. In the meantime, keeping you updated: we’re in Austin from March 14-19 for SXSW. Events we’re hosting (and RSVPs) are below. If you want to grab a coffee, respond to this email, and we’ll make it happen.

Arek and Ethan 🦄

Sunday, March 15 (Austin): We’re hosting an emerging managers + founding GPs dinner for 25 VCs at SXSW with Mercury and Sydecar. RSVP here.
??? (Austin): 👀

Resolve AI is building an AI companion for production code. Typically, engineers spend a nontrivial amount of time on remediating errors with code after it is deployed. When issues or incidents arise, they receive pager notifications, often at odd hours, and respond by diagnosing what went wrong and shipping a fix. Resolve AI detects and triages production failures early, investigates potential root causes, and suggests solutions to remediate such incidents. By focusing on reducing firefighting, it aims to save companies money and time by addressing alerts proactively and freeing engineering capacity.

Check it out: resolve.ai

The company’s pricing model varies by a customer’s scale and specific use case. It offers task-based pricing on an annual basis, where a task could be a chat with Resolve AI, an investigation session, or another action. In other words, this means customers pay a certain sum to use Resolve AI for a number of tasks per month.

Launched and raised a $35 million seed round led by Greylock Partners in the fall of 2024
Raised Series A of $125 million at a $1 billion valuation, led by Lightspeed Venture Partners, in February 2026
Customers running Resolve AI in production include Coinbase, DoorDash, Salesforce, MongoDB, and Zscaler

Resolve AI is the third company CEO Spiros Xanthos and CTO Mayank Agarwal have built together. After meeting over 20 years ago in graduate school at the University of Illinois Urbana-Champaign, the two founded Log Insight, a tool designed to analyze machine-generated log data, which was acquired by VMware in 2012. Their next company, Omnition, was a leading contributor to OpenTelemetry and focused on observability tools like distributed tracing and improving monitoring across microservice applications before being acquired by Splunk in 2019.

While at Splunk, the two frequently saw friction around alerting and incidents in production. Drawing on their broader industry experience, they felt the general inefficiency and pain around engineers on call in war rooms trying to assess and remediate issues, but it wasn’t until the recent improvement in LLM and AI capabilities that the duo decided they could finally tackle this problem effectively.

From there, the founders built Resolve AI and launched in late 2024.

Fixing code after it's deployed to production is an inevitable, and inefficient, process. Most engineers are more interested in writing and shipping code than they are in being on call, handling incidents, and resolving errors. On top of that, in some companies, the process of investigating and fixing an issue can involve many different people, causing a general back-and-forth.

Here’s what normally happens: incident alerts come in, suggesting that something is broken. Then, an army of engineers hops on a call, determines what went wrong, and discusses how to remediate the situation. This is nobody’s definition of fun, but everybody involved understands that there's a time sensitivity to troubleshooting and addressing issues that, if unresolved, have serious downstream consequences.

Whether you’re an engineer with on-call horror stories or not, that’s a problem that’s pretty universal, painful, and important.

Resolve AI addresses that by leveraging AI to better handle both alert investigation and incident remediation. Built on a multi-agent system, it uses source code, infrastructure, and observability tools for troubleshooting, working around the clock to respond to every alert. Effectively, it provides AI software reliability engineers who are always available and able to look into any issue, big or small, that gets flagged in production, working 24/7 the way a human can’t.

From initial investigation to using codebase-specific context to provide a root cause, Resolve AI will also provide proper documentation and a recommendation going forward to address the issue. In other words, for engineers, more tedious tasks that don’t require problem-solving (like documentation and messaging) are simplified. And for the truly complex production errors, Resolve AI provides a starting point on problem context and potential solutions, preventing the need to start from scratch.

An example of an interaction with Resolve AI to solve a production deadlock.

While Resolve AI isn’t the only company in this space, it has two key advantages. First, its team has extensive experience in incident management and enterprise software resilience, gained through their two earlier startups and time at Splunk. Secondly, it uses that expertise to develop something more than a general wrapper around an LLM. Instead, it builds AI models and features specifically trained on enterprise codebases, which are often highly unique and context-specific. And its early momentum, demonstrated through a long list of reputable enterprise customers, suggests Resolve AI is not only solving the problem but doing so very effectively.

After speaking with the team directly, we believe the founder fit here is as strong as it gets. The two founders not only have extensive experience with the pain points of this space, but with over 20 years of shared rapport and multiple successful startup exits, Agarwal and Xanthos understand their individual and combined strengths. That permeates through the rest of the company. For a startup still in its early stages, that chemistry and expertise between founders goes a long way.

Resolve AI wants to solve how organizations approach on-call, with incident remediation as a more collaborative activity between AI agents and engineers and investigations remaining a more autonomous, AI-only pursuit. With a deep expertise in the problem space, early fundraising traction that has already hit unicorn status, and an incredibly strong founding team, Resolve AI is poised to be the premier solution that simplifies engineers' lives and improves production reliability.

Introducing AI for PROD [Resolve AI Blog]
Interview with Greylock and Lightspeed [YouTube]
Series A Announcement [Resolve AI Blog]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

MITO

Mon, 23 Feb 2026 16:00:00 +0000

Happy Monday.

We’re excited to announce that our Unicorner team has officially grown by one! As of this week, Larissa is joining us as our first full-time hire, helping us continue building our weekly startup covers while expanding into new content. Stay tuned for more on that new content soon. 👀

In the meantime, keeping you updated: we’re in SF this week, and Austin from March 14-19 for SXSW. Events we’re hosting + RSVPs are below. If you want to grab a coffee, respond to this email, and we’ll make it happen.

Ethan and Arek 🦄

Tuesday, Feb. 24 (SF): We’re hosting a seed+ founders dinner for 20 founders with Fidelity Private Shares and Fenwick. RSVP here.
Sunday, Mar. 15 (Austin): We’re hosting an emerging managers + founding GPs dinner for 25 VCs at SXSW with Mercury and Sydecar. RSVP here.

MITO is an AI-native video creation platform for filmmakers, designers, and creative teams. The platform features six core modules: moodboards, storyboards, Figma-style team collaboration, interactive landing pages, video model integrations (Nano Banana, Kling AI, Veo, ElevenLabs, and more), and an AI agent that can make many actions on the user’s behalf. Together, these make up every step of the workflow from concept to timeline editing. In addition to its product, the company organizes MITO Films, a production service, and MITO Universe, which provides creatives a space to exchange ideas and assets.

Check it out: mito.ai

MITO is available with a freemium model. Its free tier extends 4,000 credits for exploration. Its Pro plan, available for $24 per seat, provides 20,000 credits and supports five minutes of high-quality video. The Studio plan, designed for high-volume production with priority support and extended credit rollovers, is available for $72 per seat with 80,000 credits. For large organizations, the Enterprise plan brings custom pricing with unified billing and enhanced security.

Raised $4.5 million in a pre-seed round led by Lightspeed Venture Partners, with participation from firms including Sequoia Capital, K Fund, Kibo Ventures, and Everywhere Ventures, alongside a group of angel investors
Officially launched three weeks ago in January 2026
Gained access to Seedance 2.0, ByteDance’s newest AI video creation tool that is currently limited to a select group of companies

MITO is scaling its team and hiring across Product, Marketing, Engineering, and Operations. If you are interested, reach out to us by replying directly to this email!

Taking a look at MITO’s team explains its "by filmmakers, for filmmakers" slogan.

MITO’s CCO, Arantxa Barcia, is an art director and a filmmaker with a background in fashion and advertising at companies like MIRAVEO, with specialized experience in content creation for creators and brands. Barcia holds two master’s degrees, one in digital marketing and a second in commercial management and marketing in the textile and fashion sector.

At the same time, CPO Danny Saltaren is a National Design Award 2021 recipient in the category of Young Designers by Spain’s Ministry of Science, Innovation, and Universities. Saltaren previously co-founded a digital growth studio and a product/brand studio.

Last, and definitely not least, is MITO’s CEO, Iñaki Berenguer, a serial entrepreneur who serves as a managing partner at LifeX Ventures, a $100 million fund investing at the intersection of AI and science. He previously founded three companies that were acquired, and holds an MBA from MIT and a PhD in engineering from the University of Cambridge.

The year 2023 marked a significant leap forward in general purpose AI video generation, with tools like Runway Gen-1 enabling the creation of short videos based on images and prompts. These early models also gave us some bizarre internet art, including the absolutely nightmare “Will Smith eating spaghetti” meme generated using Modelscope’s text2video model. Although the internet scoffed at these early AI video creations, fast-forward to just over two years later, and we’ve seen remarkable progress. The Will Smith meme even became a benchmark to analyze, year after year, the evolution of AI-video generation tools.

Despite the rapid improvement, most AI video tools remain one-shot solutions. This means you type a prompt, get a clip, and if it's 90% perfect, you discard it because it’s not easy to fix the remaining 10%. This can be addressed with standard video editing software, but there’s no connection between them and video generation models. In fact, video creatives have always faced unnecessarily complex, disjointed, and expensive workflows, even with legacy, pre-AI software.

MITO solves this with an infinite editing canvas. Imagine a Figma for video, where filmmakers can manage content like moodboards, storyboards, and scripts in one place. It is completely AI-native, meaning users can create everything directly on the platform.

Its product also stands out by integrating multiple leading AI models like Nano Banana, Kling AI, Veo 3, ElevenLabs, and more. This makes MITO future-proof, since it isn’t reliant on any single model and instead derives its value from bringing the best services together in one place. Moreover, it means creators don’t have to worry about juggling different tools and subscriptions, nor spend time keeping up with the newest and best models.

A demo of MITO’s collaborative storyboard feature

In essence, its value proposition is offering big studio infrastructure to small creative teams, giving them access to the same high-end orchestration tools previously reserved for Disney or Pixar.

Zooming out, the global film and video production market was valued at $316.3 billion in 2026 and is expected to reach $443.6 billion by 2035. Meanwhile, the digital ad spending market size represents a $600 billion market and is projected to more than double over the next decade.

It’s not a surprise then that the advertising world was quick to implement AI video generation tools, with popular commercial AI productions like the Toys “R” Us AI origin story or the AI-generated ad that aired during the NBA Finals.

However, with AI slop and surrealist images everywhere (the Facebook feed is there to prove it), storytelling has become the most important skill today for brands and companies. LinkedIn job postings for "storyteller" actually doubled in 2025.

Given just how many industries rely on storytelling and visual media, and how MITO is enabling brands and small teams to focus on that rather than budget and technical constraints, we believe MITO has the potential to be a blockbuster company that changes the next era of professional storytelling.

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Lotus Health AI

Mon, 16 Feb 2026 16:00:15 +0000

Happy Monday.

Free healthcare, you say?

It isn’t clickbait, we promise! Lotus Health AI wants to keep its platform free for users. It’s promising, but there’s a lot the team needs to navigate to make it achievable in the long run.

We have a lot of travel ahead of us. We’ll be in SF all of next week, and Austin from March 14-19 for SXSW. Events we’re hosting + RSVPs are below. If you want to grab a coffee, respond to this email, and we’ll make it happen.

Arek and Ethan 🦄

Tuesday, Feb. 24 (SF): We’re hosting a seed+ founders dinner for 20 founders with Fidelity Private Shares and Fenwick. RSVP here.
Sunday, March 15 (Austin): We’re hosting an emerging managers and founding GPs dinner for 25 VCs at SXSW with Mercury and Sydecar. RSVP here.

Lotus Health AI is building an AI doctor, available through an iOS app. It combines peer-reviewed training data with patient information like medical records and wearable data to build a personalized primary care doctor that can provide diagnoses, prescriptions, and more. After determining the appropriate action, the model defers to certified clinicians for final approval.

Check it out: lotus.ai

Lotus’s business model is perhaps its most ambitious feature. The company is pursuing a model where it can keep its product free for patients, which may include sponsored content or subscriptions.

Raised $35 million Series A co-led by Kleiner Perkins and CRV
Created an AI model that unifies data from millions of peer-reviewed studies with patient medical data

KJ Dhaliwal, CEO of Lotus, has a background as a founder and engineer. Prior to Lotus, he founded Dil Mil, the largest and top-grossing South Asian dating app (which was acquired in 2019 for $50 million) and was a member of the founding team of Union Metrics (which was acquired by TrendKite). He pursued an education in engineering, graduating from UPenn and Drexel University.

At the same time, Dhaliwal has thought of the inefficiencies of the healthcare system since acting as a medical translator for his parents during his childhood. His interest in the problem space, combined with the innovation of LLMs and his background, led him to Lotus Health AI, a company with a bold vision for healthcare in the U.S.

Saying the healthcare industry in the U.S. needs a revamp is a cold take. And it has been one for a while.

At the same time, AI is becoming a trusted companion for everything from double-checking emails to making life-changing decisions. Consulting AI for medical advice is just one component of this larger trend, one that OpenAI has tackled with ChatGPT Health, as a complement rather than a replacement.

Lotus’s first value add to users is its custom AI doctor. Like ChatGPT, Lotus integrates with users’ medical information, like medical records and more trendy data like wearable data to answer patient-specific medical questions.

A sample conversation with Lotus AI (refilling prescriptions)

Unlike ChatGPT, Lotus provides diagnoses, prescriptions, specialist referrals, and more. To accomplish this, it works with certified clinicians to provide final approval. On the patient side, this improves reliability and trust in the product. And on the clinician side, it improves the throughput of patients seen by a claimed >10x.

If that wasn’t enough, the company is providing its service to patients free of charge. While it’s early to say what the company’s business model at scale will be, it aims to continue offering its product to patients for free. With this, the company goes from a mere technical innovation to a viable solution to one of America’s most pressing problems. If it makes this model work, it can change the landscape of healthcare, and becoming a Unicorn would be the least of its accomplishments.

Lotus Health nabs $35M for AI doctor that sees patients for free [TechCrunch]
Lotus Just Raised $41M [Lotus]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Arena

Mon, 09 Feb 2026 16:00:27 +0000

Happy Monday.

It’s post Superbowl weekend, y’all. While you were watching the Patriots try to remember that they’re playing football, we were watching all the ads.

On that note, if you think the Anthropic vs OpenAI beef is getting a little too personal, this week’s startup might provide a more objective perspective on which LLMs are performing better!

Ethan and Arek 🦄

Feb. 19 (SF): Our partners at Lightfield are hosting a live masterclass on founder sales. If you’re a founder or early-stage GTM operator, this one is probably for you.
Feb. 24 (SF): We’re hosting a seed+ founders dinner with Fidelity Private Shares and Fenwick. RSVPs open next Monday.
Feb. 26 (SF): We’re hosting a founder fireside with a surprise guest, in partnership with Fidelity Private Shares and Intercom. RSVPs (and the guest announcement) in next Monday’s newsletter. 👀

Arena (formerly LMArena) is an online platform that allows consumers to view and compare the performance of frontier AI models. With an interface familiar to anyone who has used an AI chatbot like ChatGPT, it allows consumers to chat with the best AI chatbot, including ones they don’t actively have a subscription to. Users can use “Max,” a model router that dynamically directs queries to the most appropriate AI model. The company additionally provides users with two responses they can choose between, essentially crowdsourcing the data to rank models. These analytics can subsequently be provided to AI labs to refine and align their models.

Check it out: arena.ai

Arena sells to enterprises like AI labs looking to improve model performance.

Raised $150 million Series A from a16z, Kleiner Perkins, Lightspeed Venture Partners, Felicis, Laude Ventures, LDV Partners, The House Fund, and US Investments
Grew community by over 25x with adoption by AI labs
Boasts over 50 million votes and 400 model evaluations

Arena is founded by a true powerhouse of a team.

Anastasios Angelopoulos, Arena’s CEO, formerly worked at Google DeepMind. He received his bachelor’s in electrical engineering from Stanford and his PhD in machine learning and computer vision from Berkeley.

Arena’s CTO, Wei-Lin Chiang, also has impressive accolades, including internships at Amazon, Google, Alibaba, and Microsoft. He earned both a bachelor’s and master’s from National Taiwan University and is currently pursuing his PhD in EECS at Berkeley.

Last, but definitely not least, is Ian Stoica, a serial founder. He founded Databricks and Anyscale, additionally serving as the CEO for the former. He graduated with his PhD from Carnegie Mellon.

What began as a research project is now a trusted resource on AI model performance worth $1.7 billion.

OpenAI’s ChatGPT was a first of its kind AI chatbot when it was released in 2022. Each successive feature, like image generation and internet search, were quick hits with the public, and were swiftly adopted by competitors. Those companies even introduced features of their own.

Today, however, the common subset of features among LLMs has grown as the frontier labs have shifted their focus toward more urgent competition in fields like agents. Despite this, chatbots remain the most popular way of interacting with AI, and we anticipate this to be the case even as trendy applications like agents see increased adoption.

But while the feature set of these LLMs become interchangeable, their performance does not. Just this past Friday saw the release of two new models, Opus 4.6 and GPT-5.3-Codex. While OpenAI’s model is currently not available through a chatbot interface, the competing releases more importantly showcase the ongoing struggle to claim the top spot for performance among a number of categories.

A recent model ranking by Arena

This raises a number of questions. On the enterprise side, how do these companies measure the performance of their models? Second, how do they compare model performance with those of competitors? And third, on the consumer side, how do users of AI chatbots choose the best model for a task when models are announced so frequently (and with confusing names like GPT-5.3-Codex, too)?

Each of these is a customer pain point addressed by Arena. Arena removes the need for users to pick a model for a task, at a price that beats a standard ChatGPT Plus subscription, while frontier labs have a method of gauging their models from a source that is as close to their ICP as possible.

The team, too, is nothing short of impressive, with an equally outstanding list of investors to complement it. Altogether, Arena is a strong team building a useful product with a clever model that provides both parties an improved experience, and it’s one we only see growing in the short and long term. In our view, that’s a winning combination.

LMArena Raises $150M in Series A Funding [FinSMEs]
LMArena lands $1.7B valuation four months after launching its product [TechCrunch]
Introducing Max [Arena blog]
Fueling the World’s Most Trusted AI Evaluation Platform [Arena blog]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.

Aisepedia

Mon, 02 Feb 2026 16:01:18 +0000

Happy Monday.

We had an incredible group of founders join us in SF last week, hosting alongside our partners at Fidelity Private Shares and Lightfield.

Wednesday’s founder dinner

Henri Liriani, co-founder of Lightfield, was there as a surprise guest and shared his founder story during the evening (including pivoting from his last company, Tome).

Stay tuned for more programming coming up in February!

Arek and Ethan 🦄

Aisepedia is building a product marketing platform that combines core marketing functions such as market research, segmentation, positioning, messaging, and enablement. Designed specifically for B2B product marketing managers, Aisepedia serves as the first dedicated Product Marketing Environment. Using this environment, the entire process from start (research and data sourcing) to finish (presentation and positioning) is transformed into a single cohesive workflow.

Check it out: aisepedia.com

The Aisepedia team is considering a usage-based model for its platform. Companies would pay for credits that can be used for different features, allowing pricing to scale with value generation.

Raised close to $1 million in pre-seed funding
Launched its product to 100 beta testers in January 2026
Working with 13 design partners, including Treasure Data, a data management and customer data platform; SmartBear, a developer of quality code tooling; and Centivo, an affordable healthcare company

The team is currently offering beta testing for its product!

Vishwa Vijoyendra Narayan comes from a family of entrepreneurs. After completing his education in electronics and communications engineering at Motilal Nehru National Institute Of Technology, and later receiving a post graduate diploma in management from IIT Bangalore, he pursued a career in investment banking. However, given that the idea of building something of his own was ingrained in him from a young age, and his newly acquired technical background, he found himself leaving IB for entrepreneurship.

In 2016, Narayan launched his first company, EnCloudEn. With a goal of helping businesses run their IT systems more easily, its core product combined servers, storage, and networking into one system with the goal of making it easier for companies to manage private clouds and virtual machines. In 2021, EnCloudEn was acquired by Quantam, a NASDAQ-listed tech company with a mission to make video and other unstructured data smarter.

During his time at EnCloudEn, Narayan worked with product management and marketing. He noticed companies had difficulty viewing their product through the lens of their customers. Putting himself in the buyer’s shoes, and applying a similar, all-in-one approach he used with EnCloudEn, Narayan found inspiration for Aisepedia.

With Aisepedia, one platform is needed to go from research to content creation, replacing the mess of countless tabs. A product manager looking to understand a new product or competitor can simply enter a URL and Aisepedia will extract its core features and benefits. From there, the platform conducts industry research and creates audience segments. Using these segments, artifacts are created with information critical for GTM strategy, sales and marketing collateral, and ICP guides. For PMMs, Aisepedia is a cleaner, more organized alternative to raw research and brainstorming that allows them to find distinct avenues and angles for their product.

Traditionally, PMMs sit on mountains of unstructured data in the forms of PDFs and slide decks. Their job typically means spending more time searching for information rather than actually using it. After a project, these PDFs are rarely updated, typically sitting in a folder never to be used again. With a PME, the data is continuously updated and easy to find. With this time back and up-to-date information, PMMs can better connect with their customers, seeing precisely how individual ICP segments think and helping them understand what customers want.

A snapshot of Aisepedia’s segment builder

In addition to research, Aisepedia helps with content creation. It uses AI to generate blogs and marketing for large audiences. Having the help of the platform enables PMMs to create copy that resonates directly with audiences. It also translates into new angles and better positioning for their ICP, faster.

Product marketing is expected to reach $32.84 billion in market size by 2030, up from $11.49 billion valuation in 2023. By supporting PMMs through research to content creation all on one platform, Aisepedia is positioned to grow alongside this market, putting it on a path to become a Unicorn.

Aisepedia Introduces First-Ever Product Marketing Environment – AI-Powered, Built for B2B [Fox 59]
Aisepedia - Company Profile [Tracxn]
Product Marketing Needs an Environment — Aisepedia’s Beta Shows Why [PR Urgent]
Product Marketing Can’t Scale on Activities Alone [Aisepedia Blog]

ICYMI: We recently announced our official investor syndicate. Check out the announcement post for more details on how to invest in our deals.