My Honest Review of Google’s AI Strategy After I/O

I went to Google I/O earlier this week and want to share my thought about what I think is working and what isn’t with Google’s AI strategy.

**📌 Watch my 15-min video review with examples or continue reading **below: https://www.youtube.com/watch?v=zvMfA1VtUi0

What’s working is Gemini is growing fast. The Gemini app now has 900 million monthly users, second only to ChatGPT, and Google is also shipping a ton of new products.

But the latter is part of the issue:

There are too many AI products

Look at all the AI products that Google launched at I/O. At first glance, this seems like a flex, but I think there’s a downside. Google is launching so many AI products that users don’t know where to start anymore:

Nathan Clark@nathanclark_·May 20it’s in gemini, just create it in ai studio. oh, that’s for your personal google one account. for workspace you need gemini business. no, not gemini advanced, that’s ai pro now. unless you need ai ultra. oh agents? you do that in spark actually. no, not gemini api managed agents,Show more5112.4K18K1.5M

Between Gemini, AI Studio, Antigravity, Spark, Flow, Stitch, Pomelli, and a dozen other names, it’s getting really confusing for consumers and enterprises which product to use for what.

There are 3 AI races that matter

Instead of launching everything everywhere, I think Google should focus on winning three AI races:

• The race to evolve chat into a personal agent

• The race to build a super app for coding + knowledge work

• The race to expand beyond text to multimodal

Let’s walk through each one.

Race 1: Evolving chat into a personal agent

I think the AI chat era is coming to an end. People don’t want AI that just replies in chat, they want AI that can actually help them get work done.

That’s why I’m convinced AI personal agents are going to be a massive market. Everyone wants a personal chief of staff, even if they never plan to vibe code anything.

Here’s my mental model of the personal agent landscape:

On one end, you have OpenClaw and Hermes. These agents live in your messaging apps, are fully customizable, and pioneered this whole category. I use Hermes daily for emails, calendar, weekly reports, and more.

In the middle, you have Codex and Claude Code. These products are backed by near trillion-dollar companies and they’re rapidly adding personal agent features. But they still feel like coding tools first.

And then there’s Google. The fact is that Google already has all of my personal context. My emails are in Gmail. My calendar is in Google Calendar. My documents are in Google Docs. And all of it lives in Google Drive.

But for the longest time, the Gemini app couldn’t even edit a Google Doc, which was really frustrating.

That’s why Spark is the launch I’m most excited about from I/O.

Building a personal, powerful, and proactive agent in Spark

Google@Google·May 20Introducing Gemini Spark

It’s your 24/7 personal AI agent that helps you navigate your digital life, taking action on your behalf, and under your direction.

It runs on Gemini 3.5 and is built on @Antigravity, so it can perform long-running tasks easily in the background.Show more2551.1K5.8K1.6M

Google’s vision is to turn Gemini into a personal, proactive, and powerful agent through Spark. Let’s break down what each one means:

• Personal means understanding you through Gmail, Calendar, Workspace, Drive, and other apps.

• Proactive means telling you what matters, like the new Daily Brief feature that surfaces what to do across Google’s other apps.

• Powerful means being able to use Google’s apps but also any third-party API or MCP.

I particularly that Spark runs in the cloud on a virtual machine, so you don’t have to keep your computer open to use it. One thing Google should watch out for is to:

Let users decide how much control to give the agent

I had a great chat with Chris, the Head of Product for the Gemini app. I asked him when we can expect to hook up Spark to any API or MCP, which I can already do with OpenClaw, Codex, and Claude Code today.

His answer was that for any write action, they should probably ask the user for approval first.

I get it, Gemini has 900 million users and they don’t want people accidentally deleting all their files.

But I think this is playing it too safe.

Google should let users decide how much control to give to their agents, whether that’s asking for permission each time or just bypassing all permissions. When I work with Codex and Claude Code, I basically always bypass all permissions. These agents are smart enough now to not do the wrong thing, and they’re only going to get better over time.

The bottom line is that Google cannot afford to lose the personal agent race.

Race 2: Coding and knowledge work

Now let’s talk about coding and knowledge work.

Google is playing catch-up on coding. The AI-native builders I know have largely switched to Codex because the rate limits are generous, the app is great, and GPT-5.5 is arguably the best coding model right now.

Meanwhile, enterprises have largely switched to Claude Code because Anthropic has done an incredible job riding the hype cycle and getting adoption.

So where does that leave Google? Let’s talk about the company’s new model and harness.

The model: Gemini 3.5 Flash

First, Gemini 3.5 Flash looks like a great model on these benchmarks. It’s certainly Google’s best coding model yet.

However, pricing has also gone up, although it’s still cheaper than GPT-5.5 and Opus 4.7:

I think this actually matters because enterprises are running out of budget with expensive frontier models. They’re looking for “just good enough” and cheaper models to do the majority of their agentic work.

The harness: Antigravity

Now let’s talk about the harness. I tried the new Antigravity app and it feels pretty slick, but it also looks very similar to the Codex and Claude Code harnesses where you have a left panel to talk to your agents.

I think a harness like this is good for individual users talking to agents, but it doesn’t really work well for teams or organizations. I was hoping to see more innovation here.

Build a super app vs. add chat to every product

I also think Google has too many harnesses. For example, I don’t understand why Stitch, Google’s AI design tool, is a completely separate product from Antigravity. When I build a product, I want to plan, design, then code all using one tool. I shouldn’t have to switch between three different Google apps to do that.

OpenAI and Anthropic are both building super apps where one tool can handle coding, design, and knowledge work. Google should make Antigravity that super app.

This take might be controversial since Google is also adding AI chat to Docs, Slides, Sheets, and all of its other knowledge work products.

But I think the future is that we’ll all just interact with our personal agent and super app to get both coding and knowledge work done. Maybe we’ll go into these other apps to manually tweak stuff, but the single agent and app will do most of the work.

Antigravity needs to live up to that expectation. It needs to be incredible.

Race 3: From text to multimodal

Okay, I’ve been a bit critical about Google’s AI coding efforts, so let me end with where I think Google is genuinely ahead: multimodal AI.

Unless they really screw something up, I feel like Google is going to win consumer AI. It’s the only US lab that’s actually building competitive video models, and consumers love video. After all, TikTok and YouTube are far more popular than any text-based platform. The only real competition Google has in video right now is Seedance and other Chinese video models that don’t seem to care about copyright.

I’m also really excited for the new Omni model, which will let you take any input to generate any type of output, whether that’s text, images, audio, or video.

But even here, I think Google has too many separate products. For example:

Why isn’t Flow part of Gemini?

Flow is actually Google’s best product to generate images and videos. You can make some amazing scenes with it. But does it really have to be a separate product? Why isn’t it just part of the default Gemini app experience?

Google Flow@FlowbyGoogle·May 20Introducing your agent and the all-new Gemini Omni model in Google Flow. #GoogleIO

Think of your agent as a creative partner. It reasons through complex tasks and helps you brainstorm, create, and edit, all while under your direction.

Gemini Omni is where Gemini’sShow more140137746128K

Another pet peeve of mine: I think the number one use case for image and video editing is family photos. But Google’s safety restrictions don’t let me edit images or videos of my kids with AI. I get why this isn’t allowed, but that’s still my number one use case as a parent, so the safeguards feel too restrictive right now.

I’m optimistic about Google’s culture

I want to end on what I think Google is doing genuinely well: the culture inside the Gemini team.

Josh Woodward is probably my favorite executive at Google, possibly at any company. He keeps saying things that I believe to my bones:

• “Just try a lot and build to learn.”

• “We only have a 90-day roadmap and maybe if we’re lucky it’s 120 days.”

• “I don’t know if we’ll ever go back to one-year roadmaps. I haven’t been working on one of those in 5 years.”

Velocity over planning. Prototypes over decks. Chris, the Gemini VP who reports to Josh, told me his team caps PRDs at one page and runs meetings off Studio prototypes instead of mocks. That’s what you need to win in this space.

Chris and I

Chris and I

3 AI races that Google needs to win

So to summarize, I think Google needs to win 3 AI races (but using AI to cure cancer will be great too obviously🙂):

• Chat → personal agent. Google has the data, the products, the model, and now Spark. But they need to trust users by giving them more powerful capabilities (e.g., 3rd party APIs, computer use) faster in Spark.

• Coding → knowledge work. Google is behind on coding, but Antigravity is the right bet. Consolidate around it, build the super app, and extend it into knowledge work. Don’t just add chat windows to every Google product. Claude and ChatGPT should not have better Google Workspace functionality than Gemini.

• Text → multimodal. This is the one area where Google is genuinely ahead. Their video models are the best, and Google owns YouTube. I’m excited for Omni and they have a real shot to win consumer AI off video alone.

I’m rooting for Google. The talent is there, the data is there, and the infrastructure is there. Josh and his team are building the right culture.

They just need to focus.