Google’s Promising Us AGI

But what stuck with me was the Universal Shopping Cart

Last Tuesday, Google held its annual I/O conference, and I’m one of those people who watch the entire keynote. I came away both overwhelmed and underwhelmed. The new models are genuinely impressive, with real agentic capability and a new fast-and-cheap model, Gemini 3.5 Flash, that does frontier-level work at a fraction of the cost. Charts sped by showing AI usage exploding across the Google ecosystem. The technology is real, and it is improving ahead of schedule. What frustrated me was not the technology. It was the way Google chose to show us what they want us to use it for.

A handful of lines also stood out, particularly from Sundar Pichai, Google’s CEO, and Demis Hassabis, the CEO of Google DeepMind and the company’s resident visionary. Pichai casually folded in two examples of how people are using AI: students studying for final exams, and his own use of it to “make sense of my parents’ doctor visits.” Later, Hassabis stated, essentially as fact, that we are just a “few years away” from AGI. I know these conferences are victory laps, with their obligatory pauses for applause, like a State of the Union address. Even so, Pichai reached for two of the most contested uses of AI in the culture right now - student work and medical diagnosis - and presented both as if the debate were over and decided in their favor. Hassabis has moved his AGI timeline up to the point where he’ll say it flatly, on a stage he knows the world is watching. Each of these lines struck me wrong. But it was the demos where they really lost me.

A year ago I wrote about a significant development most people missed that rolled out at the same conference, and we’ve all been living with the consequences ever since. AI Mode, built directly into the search bar, is now just part of the Google experience. This made AI unavoidable for the billions of people who use Google search, including our students. This year, Google is clearly doubling down on its AI infrastructure investment. To what end?

All That Intelligence, for This?

The demos, one after another, were mostly small in a way that’s oddly hard to describe. Gemini Spark, the company’s shiny new personal agent, was put to work organizing a block party. In Josh Woodward’s demo it tracked down RSVPs, updated a Google Sheet, drafted emails to the neighbors who hadn’t replied, and built a hype deck in Slides, all of it running in the background. A second presenter, Robby Stein, showed something different: generative UI in Search, where it built a custom app on the fly, in this case a weekend planner for his family, which he then shared with his wife for her seal of approval. A little later came the universal shopping cart that follows you across the web, and that was the clincher for me. All that money, all that intelligence, and it was mostly pitched as a better way to throw a barbecue and a faster way to check out after an online shopping spree.

The promise of all this is that AI does things for you. The demos kept showing me how AI could build things, even directly within the search box.¹ I laughed at the weekend planner, because I know exactly what would happen if I built an app to plan our weekend and sent it to my wife. She would probably throw her phone at me. Who actually wants this? When I use Google Maps to find a coffee shop, I want the coffee shop, not to spend time fiddling with a custom-built coffee-shop-finding dashboard I can come back to later.

Over and over, they handed me solutions to problems I don’t have. So much of it felt less like help and more like homework, an endless proliferation of apps and dashboards for things families have managed on their own for decades. Google’s bet is that making these will become so fast and so frictionless that we’ll reach for them without thinking. Maybe they’re right. I’m skeptical.²

It’s Shopping All the Way Down

The part that left me most cynical is how obviously Google wants us to shop more. People already shop across Google more than a billion times a day, and the company intends to make that even more effortless. The new shopping agents work around the clock. Your cart follows you everywhere, hunts for deals while you sleep, and can be authorized to buy on your behalf with you out of the loop. One slide called it “shopping with superpowers.” The commerce segment ended on the words “happy shopping.” Then, just in case the message hadn’t landed, they scored the transition to Depeche Mode’s “Just Can’t Get Enough,” hitting Gen X squarely in the solar plexus.

I’ve no doubt some people may love this. Still, a discount on something you’d never have bought without a shopping agent doesn’t count as “savings.” It’s just a more efficient way to spend more. “Shopping with superpowers” is really a demand that we all shop with superhuman discipline, which is not a thing most of us are good at. We’ve already built a world where we can want something at eleven at night and have it on the doorstep the next day. That last layer of friction, between the impulse and actually clicking Buy, was arguably the one thing protecting some of us from ourselves, and Google’s plan is to remove even that. In a spring when gas is closing in on five dollars a gallon, inflation is climbing, and consumer confidence just hit an all-time low, do young people, and the rest of us for that matter, really need to be trained to turn every passing whim into a transaction?

Glasses That Can Whisper the Answers

Near the end came the demo I’d been dreading, and the one that should scare teachers most. Google’s first audio glasses, which speak information privately into your ear, arrive this fall; the display glasses, which show it on a lens in front of your eye, are still in limited testing. The demo was kept deliberately mundane, because the general creepiness of AI headgear is the part Google would rather not dwell on. Shahram Izadi, Google’s XR lead, and Nishtha Bhatia walked through ordering a cold brew through the frames, summarizing muted texts, and turning the crowd into a cartoon on her watch. Cool, I suppose, but do consumers really need this?

For a teacher, a device built to feed you information invisibly, whispered in your ear or shown on a lens only you can see, is an assessment-integrity problem on a scale we haven't faced, and it is not waiting for Google's fall launch. You can already buy prescription glasses with AI built in. Meta has sold millions, and a student in my class had a pair last year. Can we ban prescription AI glasses? The blue book and the in-class exam, the defenses we've fallen back on this year, do not survive glasses that give students answers in private. Beyond cheating, the glasses are just the next layer of AI envelopment Google keeps working on to make it all feel inevitable, arriving at a moment when the backlash is real and a vocal public is asking everyone to slow down. The whole keynote felt tone-deaf to it.

The Argument They Skipped

The debate about AI and education is the one the keynote treated as a closed case. The references that did appear were all upbeat - Pichai's line about students prepping for final exams, and Josh Woodward's quick run through Gemini's student features, guided learning, practice tests, and a nod to NotebookLM.³ The closest the keynote came to demonstrating actual learning was Robby Stein's astrophysics moment, where Search built a custom interactive visual explaining how black holes affect space-time. Even that was staged as an exciting feature of what students can do. Yet the question that keeps teachers up at night, whether these tools are actually helping students learn, never came up. Google ignored the most contested issue in my world and declared it settled while the rest of us are still figuring it out. It won the argument by declining to have it.

After the keynote, Dan Fitzpatrick (no relation), a highly sought-after voice in the AI-and-education world, sat down with Chris Phillips, the man who runs education at Google, and came away with a much more nuanced, teacher-centered case the keynote never bothered to make. His piece is worth reading in full.

Two of his takeaways landed, for different reasons. The first described one of the most impressive demos in which Antigravity 2.0, Google's revamped agentic coding platform, built a working operating system from scratch, mostly on its own, for under a thousand dollars. Building something that complex for that little has real ramifications for developers, but it is not likely to impact teachers any time soon. It is just another example of the gap between what the frontier can do and what the 2026 classroom is actually concerned with.

His larger point is that Gemini 3.5 Flash is now good enough to do almost everything schools are asking for, and the new price point will lower the cost barrier keeping AI out of districts willing to make the switch but unable to commit the resources. I suspect it will make AI in schools more likely, not less.

Though Dan’s piece might be interpreted as more bullish than I am on where AI in schools is headed, it did not offset my visceral and mostly negative reaction to the keynote itself.

Even though it’s clear from his reporting that Google has a thoughtful account of what its tools do for learning, it’s telling that it chose not to put that on the stage.

The Promise vs. the Product

The keynote opened on a sizzle reel of possibility - a young boy on a tablet doing something undefined but clearly interesting with AI, a teacher mid-music-lesson, a few small-business owners, over a voiceover about solving disease and “making something that matters.” There were legitimate use cases: a Korean immigrant who used these tools to build a hiring platform for her neighbors, and creative AI tools for musicians and filmmakers.⁴ It closed on Hassabis invoking the “foothills of the singularity.”

Hassabis is the AI paradox in one man. He came out early to announce that AGI is just a few years away, and he returned at the very end as the closer, touting AlphaFold, Isomorphic Labs chasing the goal of one day “solving all disease,” and WeatherNext, the forecasting model that called Hurricane Melissa’s path days early, in time to get people out of the way. These are all real ways AI has advanced scientific development. He promises a “new golden age of scientific discovery” that improves lives everywhere. The demo that ran just before that close was the cold brew, ordered through a pair of glasses. From a cold brew to the singularity in a single cut. Everything between his two appearances, the apps, the shopping, the glasses, was the version of AI its skeptics fear most. Which future is Google actually building?

To What End?

Pichai mentioned that Google expects to spend something like $180 to $190 billion on infrastructure this year, roughly six times what it spent in 2022. Spark, the advanced agentic tool covered in the keynote that works while your laptop is closed, requires a Google AI Ultra plan, which runs between $100 and $200 a month. What's free for everyone is the new search box, the on-the-fly generative results, and the cart. Google would tell you the cart is what pays for the moonshot. Maybe. AGI is held out as a tantalizing promise, but the cart is what turns a billion shopping searches a day into the money that pays for it.

There were genuine standouts - the aforementioned Antigravity build and Gemini Omni, the new video model, were both impressive. With Google Flow’s editing tools I could see both the fear and the possibility in it for anyone who fools around with photos, film, and music. We will likely absorb Spark, the in-search app builds, and the universal cart the same way we absorbed AI Overviews, without too much fanfare. Maybe I’m just grumpy, but a lot of it felt more like gimmicks than upgrades.

This all tells me that with agentic AI now baked into how we use Google, the fantasy of “banning” AI in schools should be over, if it was ever a serious position to begin with. Yet we still barely know whether, or how, to teach with these tools at all. For the first time, as someone usually curious and fascinated by this stuff, I’m feeling more AI fatigue than excitement. Maybe I am just suffering from AI exhaustion.

Still, what stuck with me most was the shopping. The direction Google is pointing, as best I can tell, is to get me to buy more, build disposable apps to help me do it, and surrender to a world where every experience is something to digitize, optimize, and monetize. AGI promised something bigger than this. Do I really need a customized shopping experience, or bespoke apps for using the web? And do I want a pair of glasses talking into my ear or digitally interpreting my world so I never have to engage with it? The 2026 keynote was a tale of two futures, and the one that feels likeliest after watching it is the one I want least.

---

Connect With Me

Beyond this newsletter, I work directly with schools, educators, and other institutions about pedagogical questions raised by AI. Take a look at my website and reach out - I’d love to hear what you’re working on.

1. Regular readers will note the irony here. I've spent the past several months building detailed, student-facing web modules for my courses which I've written about. The difference, in my mind, is that those take real up-front planning and metacognitive work, and they're built to last. What Google demoed were instant, disposable versions of the kinds of things you might make once and likely never use again. Is this really something people were waiting for?

2. To be fair, I’ve dismissed tech advances before. I was someone who never wanted a navigation system. When in-car GPS arrived, I drove around for years without one, sure I wasn't missing anything, until I finally got a new car and now I can’t imagine driving without one.

3. I love NotebookLM. I find it useful, and one of Google's most interesting and unique contributions for using AI effectively. My overall point is that the keynote showed no awareness of the raging debate over the cost-benefit analysis of student AI use in schools.

4. Though it’s worth noting that the use of AI tools for creatives is also among the most controversial and contested areas of AI discourse. They also upgraded and showcased improvements with Google Flow Music - another area of intense reactions within the industry. Again, Google treats these cases as settled inevitability.