Google 'I' Hold the 'O'

The weeks and months leading up to the Google I/O conference are funny in the Age of AI. Everyone online seems to talk themselves into the notion that Google has fallen behind because they're seemingly more quiet than their rivals. After a slow start, they were able to catch up, so why hasn't Google kept their foot on the gas? That's the general sentiment. Then I/O hits and it showcases Google putting their pedal to the metal, zooming back in front of the pack.

At least, that was last year. This year was a bit different.

While yesterday's keynote showed the company effectively using their size and scale advantages with a wide array of announcements across a huge swath of services (many of which have billions of users – 13 of them now over 1B, as CEO Sundar Pichai highlighted front and center), it also felt a bit underwhelming, if I'm being honest. Perhaps that's just AI as a technology maturing and those of us living with it daily getting naturally numb to the incredible capabilities (as always happens with all technology). But it also feels a bit like Google may need to rethink their strategy around how they handle such roll-outs.

I'm not saying they need to ditch I/O, but I do think they'd be wise to make it less of a big public showcase event and more of an actual developer event. Google naturally followed Apple down this big keynote path, as did everyone else. And the strategy seemed to work well when Android was combating the iPhone. But the Age of AI is different. And most notably, that includes the cadence of releases.

Obviously Google doesn't fully wait for I/O each year to release what they've been cooking up over the past many months. Like every other AI lab, they're constantly pushing out new models and capabilities. But at the same time, unlike the other AI labs, Google clearly also holds some things back in order to make a splash at I/O. Only this year, the cupboard seemed relatively bare.

Don't get me wrong, a lot of what they talked about and showcased looked cool. The problem is that almost none of it actually seems ready to roll. As in out.

While they did seemingly move the entire stack to Gemini 3.5 Flash, the actual frontier model, the Pro variant, clearly isn't ready. To the point where it was only mentioned by Pichai in passing, noting that it would be coming next month. The audience didn't like this too much. And while it's undoubtedly better to let a model bake longer rather than show it off (let alone push it out) prematurely (just ask Meta to know how that goes), it sort of highlights my point: what's the point of a big, forward-facing single keynote in the Age of AI?

You either have to hold stuff back to show it off there, which is bad when your competitors aren't doing that. Or you have to move stuff forward to be able to showcase it – but not ship it – at the event, which is arguably worse.

Again, there was a lot of the latter. Nearly everything shown off on stage was either going to be coming later this year or would be rolling out in beta mode to those willing to pay for the 'Ultra' tier of Gemini.¹ If you're just a regular old Gemini user – or even a mere 'Pro' peon – there's not a lot for you to actually try today.

I'm not suggesting this will be like the don't-call-it-vaporware – but awfully vaporware-y – AI that Apple talked up at WWDC two years back. But again, Google just held a giant event to talk up a bunch of products and functionality coming soon. The other AI labs also do some previewing, but they also don't hold events at the scale of I/O. And their previews are often a result of them being either resource constrained – or they're Anthropic worried that 'Mythos' could end civilization as we know it. Again, Gemini 3.5 Pro seemingly just isn't ready.

And that's fine! But maybe hold a smaller Gemini 3.5 press conference say, next month, when it is ready?

Same with the new 'Omni' class of models. Again, the new video generation looks good, but it's also clearly a fraction of what Google is hoping to do here. This also felt a bit underbaked in terms of delivery to the public. And even the more limited subset starts by rolling out only to paid users at first.

'Gemini Spark' – how cold of Google is it to name their OpenClaw-killer after Meta's first attempt to catch up in frontier models? – will come to "trusted testers" first. Then to the 'Ultra' tier. Then to the others at some later point. It will apparently be baked into Chrome "later this summer" but probably not everywhere, and probably not for everyone.

'Android Halo' is also coming later this year, though no one seems to have any idea what that is beyond the extremely vague mention of agents by Google. (It may just be a UI tweak for Android?)

Antigravity did get a "2.0" version launched yesterday, but people seem fairly underwhelmed in the space that's arguably the most competitive and important right now in all of AI? Google perhaps should have waited for 3.5 Pro to ship it? Again, that will be next month, it seems.

There are a number of new things coming to Google Search itself – almost all of which are coming later. The generative UI? This summer. Agents and mini-apps? This summer. The new shopping capabilities? This summer.

Wait a minute. The new voice capabilities demoed for the Gemini Mac app? This summer. Google's new 'Pics' photo editor? This summer. Again, Spark in Chrome? This summer.

It almost feels like Google should have instead held an event... this summer.

The new 'Daily Brief' feature in Gemini is apparently rolling out soon, but to paid users at first. At least the new "Neural Expressive" UI seems to be actually live – and looks quite good, I might add!

How about the Gemini-powered Smart Glasses – sorry, 'Intelligent Eyewear'? These have been teased or previewed at least three or four times already, maybe more. Surely they're ready by now to meet Meta in the market, right? Right, if by "now" you mean this Fall. For the 'Audio Glasses' (yuck branding) at least. The ones with the display in the lens? That's looking like next year.

So we'll presumably see them previewed or showcased at least a few more times before they actually ship.

Look, I'm being a bit of a jerk here. My point is simply that I'm not sure how much it makes sense to use Google I/O as a showcase in the Age of AI. Things are just moving much too fast (or in some cases, too slow) for a once-a-year big event. I think it makes sense to convene your developer ecosystem in such a way, but just like with Apple's WWDC, these have morphed into being too forward-facing (with the expectations to boot). For actual products (and models), it probably doesn't even make sense to follow Apple's other Fall event (sometimes two) and/or Spring event cadence. Instead, perhaps do what OpenAI does and just convene people when you actually have something to talk about and showcase. Ideally right around when the work is actually ready to roll.

A few more things:

• Interesting that Pichai highlighted Gemini being at 900M MAU, both because that's seemingly right around ChatGPT's most recent metric, but also because like ChatGPT, they'd clearly much rather announce when they hit 1B. It has taken OpenAI quite some time to get there, presumably it won't for Google – as it should be their 14th such product soon.
• Wild that he also used the opening part of the keynote to specifically call out their CapEx spend! But that's obviously an attempt to retake such narratives and point out that unlike their competitors, Google is doing the "full stack" build-out here, namely with TPUs (though Amazon can obviously make that case too).
• The first Demis Hassabis bit started with a pretty bold bang: "AGI is just a few years away." In the past, he's been far more hedge-y there in terms of timetables. I'm sure he wouldn't be pinned down to say "3 years" but to me, "a few" reads like "3 or 4". AGI before 2030?
• Again, per above, it felt like Google wanted to come out with a big swing for Omni as their first true "World Model", but had to pull back a bit. There's definitely the subtext that Hassabis believes such models will be the key to unlocking AGI...
• The benchmarks that Google showed off for Gemini 3.5 Flash were obviously impressive, but some ratings put out there by others? Less so.
• Regardless, the big talking point after the keynote seems to be the price increase for such a model.
• While there was a lot of dunking online for Josh Woodward using an iPhone to demo Gemini's agentic capabilities on mobile, I actually think that's a great thing – showcasing they care how it works regardless of which ecosystem you're working in. (And that was obviously intentional!)
• I'm less clear if the Codex portion of the Antigravity demo was...
• The first major update to the actual Search box in 25 years is obviously a big deal, though it feels like shoving "AI Mode" in there was also a pretty big – and bold – change last year. And clearly it seems to have worked! So now this is just making permanent that change in behavior they guided...
• Hassabis came back out to talk about scientific breakthroughs (again kicking off with "AGI is on the horizon") but also clearly felt the need to address the Mythos titan in the room (without actually naming it), noting that Google has been focused on security for decades.
• Hassabis closed out the keynote, not Pichai. Interesting.

¹ Google did lower the price of the 'Ultra' tier, but it almost felt like a move made knowing how annoying restricting so much new functionality would be, and trying to throw a bone to those who will undoubtedly want to try some of this stuff now. ↩