I’ve been feeling tired lately. Tired of keeping up with new AI products and features announced every week or so. It’s been super hard to stay up-to-date with what the top 5-7 tech companies are releasing, let alone what the small AI startups are doing.

At first, I thought it was a skill issue. Everyone seems to be on top of things except me. But recently it dawned on me: it's not my skill issue, AI is just poorly designed.

Before I make my case for why AI is poorly designed, read the list of new products and features Google announced at I/O 2026:

Models

• Gemini 3.5 Flash

• Gemini 3.5 Pro

• Gemini Omni

Platforms & Developer Tools

• Google Antigravity 2.0

• Managed Agents in the Gemini API

Search

• Redesigned AI Search Box

• Information Agents

• Search Mini Apps

Shopping

• Universal Cart

Gemini App

• Gemini Spark

• Personal Intelligence

Creative Tools

• Google Flow & Flow Music (mobile apps)

• Google Pics

Workspace

• Voice capabilities in Gmail, Docs & Keep

• AI Inbox updates

Science

• Gemini for Science

Safety

• SynthID & C2PA Content Credentials expansion

This is what Google has already shipped or what it will ship in the near future. And that’s just one company. There are other companies that will do the same. And Google will surely ship more than what’s listed here within the next year.

Raymond Loewy is often called the father of industrial design. He designed a lot of iconic things, like the Coca-Cola bottle and the interiors of NASA spacecraft. Loewy coined a design principle called MAYA, or Most Advanced, Yet Acceptable.

This principle states that humans are caught between two opposite forces:

• Neophilia: A love for new things

• Neophobia: A fear of the unfamiliar

When designing a new product, designers have to cater to both of these forces. The product should have some novel angle to it, but it cannot be so novel that the users don’t feel comfortable using it.

When the world shifted from simple buttons to the iPhone's multi-touch screen, it was skeuomorphism (designing digital interfaces to look like their physical counterparts) that kept them comfortable enough to try the novel concept of multi-touch screens. Thanks to MAYA, that shift happened relatively smoothly.

The velocity of AI progress offers novelty. But it offers nothing to soothe the deep-seated human fear of the unfamiliar. AI products practically didn’t exist before 2023. And now, a mere three years later, an average user is supposed to use and accept dozens of new AI products/features/interfaces every month?

I am an average user; testing new AI products isn't my full-time job. Had that been my job, not staying up-to-date would have been considered a skill issue. But since that’s not the case, I call it a failure of tech companies to follow a well-established design principle.

I am not trying to throw shade at AI companies. They are acting this way because they are caught in an existential race. My purpose here is to tell you, the average user, that it’s okay if you are feeling exhausted with staying up-to-date. Raymond Loewy predicted our exhaustion decades ago.

Let's play a fun game.

Think of 10 random software/tech companies, visit their homepage, and search for ‘AI’.

I did the same, and here’s what I found:

1. Brave Browser

I like Brave Browser. So I started with it. Found AI at a first glance. Huh, must be a coincidence.

2. Firefox Browser

Brave’s alternative Firefox, too, couldn’t resist putting AI on its homepage.

3. Proton

I use Proton for email and VPN. Now they’re offering an AI Chatbot too.

4. Notion

"Look how they massacred my boy." — Vito Corleone, The Godfather (1972)

5. DuckDuckGo

Props to them for choosing a nice domain: duck.ai

6. Beehiiv

Beehiiv is the platform I use for creating this newsletter. And they don’t use any generic website builder, but only an original, visionary, one-in-a-million AI website builder.

Btw, I might switch to another platform because the website builder is so buggy.

7. Kit

Then I checked Beehiiv’s alternative Kit.com. Yaay, first website to not have AI on its homepage.
Should I switch? Is this a sign from the universe?

8. WordPress

Then I remembered the OG Internet website: WordPress.
That’s the second website that doesn’t feature AI on the homepage

9. Cloudflare

Talking about OG internet software companies, I went to Cloudflare next.

10. Shopify

Next AI all-star? What’s that, Shopify?

Can you guess what my point here is? 8 out of 10 random software websites I checked feature AI on their homepage.

I don’t want to over-intellectualise and over-explain my conclusion here because the conclusion is so evident. It’s the B word.

Bubble.

This week’s edition is a special one: Chai with Android 17. Nothing major happened in the AI space last week that would help individual users except for the announcement of Android 17.

Google held ‘The Android Show: I/O Edition’ last week and announced new features and products. There are some caveats to these announcements, which we’ll cover at the end. But first, let's dive into the highlights:

1. Cross-application Workflows

Currently on Android, sharing context between apps is entirely manual. For example, if I message my friend the location and time of our meeting on Signal, my friend will have to copy that from Signal and paste that to his Calendar app.

In the upcoming days, AI will be able to use context from one app and use that agentically in other apps. Google showcased a few examples, like asking AI to check your notes and pick grocery items from that, search for the same items in the shopping app, and put them in the cart.

2. Autofill: Now powered by AI

The current implementation of autofill generally uses your names, addresses, email addresses, etc. But Google announced that it will move Android autofill from pre-defined databases to something more dynamic that can get context from your documents and apps as well. For example, when filling out a web form, you will be able to ask the autofill to grab your photo from your Google Drive and paste that into the form.

3. Gemini Nano v3: SOTA on-device generative model

Not all AI-related features in Android and Chrome Browser will depend on cloud-based AI models. Some of the features will be handled by Google’s on-device generative model. The most capable of the Gemini Nano series will be Google Nano v3, which’d handle a text-to-text and image-to-text inference speed of 940 tokens/second. Now we know why Google DeepMind has been putting so much effort into smaller models.

4. Googlebooks

No, I am not talking about Google Books, which is a Google product related to digital books. Googlebooks is a new line of AI-first laptops.

The company is moving on from ChromeOS and Chromebook, and shifting to a new operating system featuring workflows designed entirely around AI. The new OS is expected to be a mix of the Android mobile stack and the ChromeOS desktop interface.

We know that Chromebooks were a good choice for students who had browser-centric workflows and could replace local files and apps with cloud files and web apps. How Googlebooks differ from this is not yet clear. Someone can argue that this is an exact parallel of what Microsoft did with its pivot to AI with Copilot-OS aka Windows 11.

5. Magic Pointer: Gemini-powered cursor

In the current AI paradigm, the user types out text or uploads a picture to an AI model to give context. But Google announced something called a Magic Pointer that replaces the current paradigm by understanding exactly what your cursor is pointing at to get relevant context.

The user will be able to turn on the feature by wiggling the cursor. Once the Magic Pointer is on, it will have the location data of the cursor, the semantic understanding of what it is pointing to, and the user's voice prompt.

This model of context providing sounds like a hybrid of Chrome and Firefox, in which we can highlight text and ask related questions, and the AI-native browsers like Comet and ChatGPT Atlas.

6. Gboard Rambler: Voice-to-text dictation

Apps like Wisprflow and Handy are quite popular and can handle voice-to-text dictation very well. Google finally remembered it has Gboard that can be used to dominate the voice-to-text market. So, it has introduced Gboard Rambler.

Rambler will not only filter out filler words, such as "um," "ah," and "you know," but it’ll also process self-correcting sentences. For example, if I say, “Let’s start tomorrow by 8 o’clock, or rather, 9 o’clock.” Google Rambler will process it as “Let’s start tomorrow by 9 o’clock.” Neat.

7. AppFunctions: MCP for mobile apps

If you’ve been following the world of agentic AI, you probably already know what Model Context Protocol or MCP is. If you don’t, it’s like an API for AI agents to look into the backend servers. Android is getting an MCP of its own, and it will be called AppFunctions.

Up until now, most AI agents interacted with apps by taking screenshots of the Graphical User Interface (GUI) and processing those screenshots. But that is a solution by brute force. AppFunctions, like MCP, will be a more reliable and elegant alternative to the fragile screen-scraping techniques.

There are more announcements, but I didn’t find them interesting.

At last comes the reality check. These announcements are mere announcements. We don’t know which of them will turn into real features and products at what point in time. And even if they do turn into reality, we don’t know if the implementation will be good enough or not. Let’s not get hyped until we get high-quality software and hardware in our hands.

“If Linux is so good, why aren’t more people using it?”

Someone once posed this interesting question on X.com. At first, the answer to this question seems obvious: if more people aren’t using Linux, maybe Linux isn’t that good. The people have spoken through their choice.

But the first answer that comes to mind often isn’t the right one. The rational part of our brain takes some time to work. DHH, the creator of Ruby on Rails, published a blog post to answer the question. He wrote:

“If exercising is so healthy, why don’t more people do it?
If reading is so educational, why don’t more people do it?
If junk food is so bad for you, why do so many people eat it?”

He continues, “It’s easier to be fat and ignorant in a world of cheap, empty calories than it is to be fit and informed. It’s hard to resist the temptation of minimal effort.” That’s it. That’s our answer in one line:

“It’s hard to resist the temptation of minimal effort.”

We can understand this from an evolutionary perspective. Up until a few generations ago, food shortages were frequent, and a lack of food meant a lack of energy to survive. The more mental and physical energy a human could save, the better their chances of survival. We are hardwired to think and do less. And this desire to think and do less leads to not choosing Linux.

But that's a discussion about OS adoption. We are more interested in AI adoption. Who’ll get adopted the most: OpenAI, Anthropic, Google, DeepSeek, or something else? Which harness will win: Claude Code, Codex, OpenCode, or Pi?

The answer, I’m afraid, might lie in the sentence we just read above: “It’s hard to resist the temptation of minimal effort.”

Big players like OpenAI, Anthropic, and Google are attracting users with subscriptions that require minimal effort. They are offering all AI-related tools under one roof. For example, if you get Google’s AI Pro Plan, you’ll get access to:

• Chatbot

• Image generation model

• Video generation model

• NotebookLM

• AI Studio

• Antigravity

• 5TB cloud storage

• Google Home Premium Standard Plan

• AI features across all other Google products

Get one Google AI subscription, and you won’t have to make an effort to find the right product for a task. Everything is available in one dashboard.

But one can argue that Google is an exception because it offers more digital products than any other company.

Well, we can also take the example of Anthropic, a company famous for not wasting effort on side products. If you get their Pro plan, you’ll get access to:

• Premium Models in chat

• Claude Code

• Claude Design

• Claude Cowork

• Claude for Microsoft 365

• Claude for Microsoft Outlook

• Claude Routines

There is an alternative path that empowers users, just like Linux does, but it takes effort. They can use  API keys for multiple open-source models through OpenRouter or OpenCode Zen. They can use open-source harnesses like Pi or OpenCode. They can use OpenDesign for AI-led design. They can use n8n for automated AI-assisted task execution. But a majority of us will probably subscribe to a bundled, closed-source, tech giant product for the same reason a majority of us use Windows and Mac. “It’s hard to resist the temptation of minimal effort.”

I am not throwing shade at people who use bundled products that offer convenience. I am just observing the human tendency to minimise effort. And if human nature stays consistent, a decade later, we’ll probably see a YouTube video titled ‘2036 is the year of Pi.’

Finally, a reliable Claude Code.

Many people moved from Claude Code to Codex last month as Claude's services were frequently going down, and people were hitting rate limits without completing any meaningful work. Finally, that might change.

Anthropic has signed a deal with SpaceX. Within a month, it will be able to use all of the computing capacity of the Colossus 1 data center. In case you’re wondering, it’s called Colossus because it’s a cluster of 220,000 NVIDIA GPUs.

Thanks to the SpaceX deal and other recent deals with compute providers, Anthropic has announced that it will double Claude Code’s five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans.

The most basic, free version of ChatGPT has gotten smarter and more reliable.

OpenAI announced GPT-5.5 Instant – its latest foundation model that hallucinates less than previous fast models. The company has decided to make GPT-5.5 the default ChatGPT model, and that’s good news for two reasons:

1) AI companies provide the baseline models (like Gemini 3 fast, Claude Sonnet 4.6, and now GPT-5.5) for free only when the inference cost is low. The more effort companies put into baseline models, the better it is for developers who want intelligent models for a lower cost. Big companies can pay high amounts, but smaller companies and developers still find SOTA models unaffordable.

Interestingly, the API cost of GPT-5.5 Instant is the same as that of GPT-5.5. The actual savings come from GPT-5.5 doing little to no reasoning, and hence using fewer tokens per task.

2) A large part of the world cannot pay a $20/month AI subscription fee. And not having access to ‘intelligence’ creates disparity. But smarter and lighter models can level the playing field. GPT-5.5 Instant is another step towards leveling the playing field.

Speaking of smart and light models, Google’s Gemma was in the news last week.

Google’s Gemma 4 AI models got ~3X faster.

Google calls its Gemma 4 series “byte for byte, the most capable open models.” So far, there is no reason to doubt Google’s claim. Last week, the series might also have become the fastest for its size.

LLMs produce one token at a time. But Google is trying to change that with a technique called Multi-Token Prediction (MTP) drafters. In this technique, a small draft model proposes multiple tokens that are validated by the main model. This improves inference speed without affecting the quality of the final output.

DeepSeek is raising funds for the first time.

It’s only been around two weeks since DeepSeek released the DeepSeek-V4 series, and the models have already become the favourite of programmers. I’ve seen dozens of posts praising the intelligence per dollar of the DeepSeek-V4 series, and none that criticize the models.

DeepSeek might soon raise its first venture capital round at a valuation of about $45 billion. There are two interesting things to notice here:

• DeepSeek has never raised money. Right now, around 90% of the company is owned by Liang Wenfeng, the CEO and founder of a Chinese quantitative hedge fund.

• Apart from the big 4-5 American AI labs, DeepSeek has impacted the world of LLMs the most. And yet, its valuation is only ~45B?

Elon Musk vs. Sam Altman: A window into the psyche of AI leaders.

You probably already know about the ongoing lawsuit between Elon Musk and Sam Altman. I don’t want to delve much into it, as the facts coming out of the proceedings are too juicy. Had Chai With AGI been a tech gossip newsletter, I would have happily given you more details.

However, there is one thing I want to mention: do read a few articles about the proceedings, read the chats and emails between tech leaders, and you will know how fallible many of these ‘leaders’ are. Certain excerpts have been shared from the diary of Greg Brockman, the president of OpenAI, that provide a clear window into the psyche of these hotshot executives.

Meta is using bone structure analysis to identify kids.

Instagram and Facebook are now using AI to analyze the bone structure of photos. If the AI detects that the user is under 13, it removes the account.

How interesting.

Children using internet services ought to be regulated by parents and guardians. But for some reason, Meta is doing that. I don’t want to bash Meta here, as there must be a certain number of accounts used by children under 13 that Meta had to develop a technology for this. I just find it interesting that a large number of parents are okay with their children under 13 having Facebook and Instagram accounts.