What the Googlers not on stage said at I/O 2026

For a first-time attendee, Google I/O’s energetic, optimistic atmosphere felt almost like a coronation.

Last year’s bets are now growth pillars because they worked. Ask Maps became the playbook for rolling out Ask YouTube. Gemini 3.5 Flash powers Antigravity — think Claude Code, but Google — and Googlers are already using it to build the features demoed on stage.

Everything shipped fast, and everything felt confident.

There was something for everyone.

• Gemini Omni, which was compared to Nano Banana but for video (I have bizarre proof).
• Smart glasses are making a comeback.
• Video game-like experiences that can be prompted and played in real time.
• Workspace can now talk documents into existence.
• Google Maps images can be turned into surrealistic fever dreams via prompting (I asked what the use case was, and it sounded more like a solution looking for a problem: Hollywood studios could forgo shooting on location?).
• I even have Gemma running on my phone so I can converse with a smaller model on an airplane. (P.S. American Airlines now has free Wi-Fi, so I’m good.)

But I haven’t even gotten to the part that’s most curious.

Gemini is becoming more like Search. Search is becoming more like Gemini.

There are now features across both products that serve the same intent: monitoring the web and proactively notifying you when something relevant appears.

In Search, it’s information agents. In Gemini, it’s Spark or Daily Brief. The overlap is obvious.

So I asked one of the product managers directly: “How are you thinking about long-term feature management and the bloat of utilities that largely overlap?”

The answer: “Right now, it’s all about velocity.”

They’re shipping relentlessly. Three other PMs behind flagship I/O features said the same thing. Every one of those features was started and shipped this year, in 2026. That was mind-blowing.

The PM added: “The way velocity is achieved is less managerial overhead.”

I took that to mean: get on the board now and figure it out later.

Once you see it, you can’t unsee it

With that framing, the rest of the day looked different. I saw plenty of impressive demos, but kept wondering: what do I actually do with these next?

I now have Gemma on my phone, but one of the developers couldn’t offer much of a day-to-day use case. I got a demo of AI Mode’s monitoring capabilities by prompting “keep me updated” and saw how the pieces connected. But when I asked a follow-up — “How will I manage these alerts alongside everything else? What happens when they go stale?” — there wasn’t an answer. Granted, it’s still a demo, but the non-answer was telling.

The second-order effects of many of these features don’t seem fully considered. It gave me the sense that engineers dogfood these models from the command line, not the front end.

One small but revealing example: as of this writing, I still can’t delete old Gemini chats in the web browser, even though I can in the Mac app.

Universal Cart: The feature that got everyone talking

One feature that came up repeatedly in conversations with both engineers and users was Universal Cart, Google’s new cross-surface shopping protocol.

When people asked what I thought, I said: “If you’re Google, you should be very excited, because if this gets adopted, you own more of the end-to-end experience. If you’re everyone else, you’re probably worried.”

That didn’t seem to concern the group I spoke with, many of whom felt oddly detached from the growing anti-AI sentiment in the U.S.

Later, I spoke with an SEO professional at a large ecommerce brand already implementing Universal Cart. When I mentioned the velocity comment, they said: “That sounds like what we experienced during implementation. It feels rushed.”

The AI content guidelines paradox

The velocity-over-oversight mindset also helps explain why Google’s AI content guidelines have been so controversial.

Four days before I/O, Google’s Search quality team told publishers to “write for humans, not AI.” Then the AI agent team took the stage and demonstrated a future where Google’s own agents browse, interpret, transact on, and generate content across the web.

If the future is increasingly AI Mode — with agents building, fetching, and acting on users’ behalf — the guidance to publishers starts to ring hollow.

Why this matters for the web ecosystem

I don’t want to diminish the work these engineers are doing. I told them that directly. As someone building products for search and for our clients, I empathize. You mostly hear criticism, not praise.

But I can’t help wondering what happens when all these overlapping features — the bloat, the inability to delete, manage, or reconcile things cleanly — become technical debt that has to be unwound. Right now, the AI playbook seems to be: feature utilization first, fix it later.

Still, I honestly respect that a company as large and established as Google is moving this fast, and I’m genuinely excited to see how some of this plays out. With their cash flow and their ability to manufacture their own TPU chips, they can afford to place multiple bets and see what sticks.

I wanted to keep talking with that PM, but we were unceremoniously kicked out of the area.

The bright spots are real

Google reported that last quarter saw an all-time high in search queries. They’re taking authentication and provenance seriously, with SynthID expanding into Search and Chrome, new adoption partners like OpenAI, and C2PA content credential verification for crawl.

Those are meaningful steps forward.

But this pace will likely create unintended consequences. My hope is that the rush to move fast doesn’t further destabilize an already-rattled web ecosystem by breaking too many things along the way.

All of this is to say: it’s an exciting time to be in search.

Dig deeper.