Hot take: Gemini 3.5 Flash is not the biggest Google I/O news

Simon Gonzalez De Cruz

2h • 💬 General discussion

Google I/O is still happening, so I’m not treating this like a final recap. This is my day-one read.

Yes, Google announced Gemini 3.5 Flash.

That is news. But it is not the part I keep thinking about.

The bigger move is that Google is building the whole workbench around the model.

Not just “here is a smarter chatbot.”

More like:

model + tools + runtime + memory + human review.

That is the shift.

Gemini 3.5 Flash is the engine. Google says it is already showing up inside the Gemini app, AI Mode in Search, Antigravity, AI Studio, Android Studio, and enterprise products.

But Antigravity is the part that feels more important to me.

Antigravity 2.0 is a desktop app for running multiple agents in parallel. There is also a CLI and an SDK. That means Google is not just shipping a model. They are shipping a place where agent work happens.

Then there is Managed Agents in the Gemini API.

That is where this gets very real for builders. Agents can run in an isolated Linux environment, use tools, execute code, browse, manage files, and resume with state intact.

That last phrase is the whole game: state intact.

Because one-off prompts are useful, but they are not where the leverage is.

The leverage is when the system can keep the thread alive.

What did we try?

What changed?

What broke?

What files matter?

What should the human review before anything ships?

That is the difference between a clever answer and an actual workflow.

Search is getting this too. AI Mode is moving to Gemini 3.5 Flash globally, and Google is talking about Search agents that can go work in the background for Pro and Ultra users.

Gemini Spark is another version of the same idea: a persistent personal agent running under your direction.

So my read is simple:

The model race is not going away.

But the next real advantage is the harness around the model.

For people who build in loops, this matters a lot.

I do not work in one clean straight line. I move between projects. I leave context in one place, pick up another thread, come back, connect two things that did not look connected at first, and somehow the whole circle starts closing at once.

A bad AI setup makes that worse.

A good one gives that pattern rails.

That is why this feels bigger than “Google launched another model.”

The model is the engine.

The harness is the leverage.

Question for you:

Where do you think the next real advantage is: better models, better tools, persistent memory, or better human review?

1 comment