Sundar Pichai walked on stage in Mountain View yesterday morning and moved more ground in two hours than the previous six months of AI announcements combined. A new frontier model. A new creative model. A persistent personal agent. An agentic shopping protocol. An agent runtime inside Search. A hardware surface. A pricing reset that halved the entry point for the top tier subscription. Top to bottom, shipped on the same Tuesday morning.
For three years the consensus said Google had fumbled the AI moment. The consensus was wrong. Google was never asleep. They were assembling. Yesterday they put the assembled thing on a stage.
I watched most of it after the fact, in the evening, having spent the day training a creative marketing team and on back to back calls. Two founders. A CMO at four. The conversations were almost interchangeable. Different sectors, different stages, same sentence underneath it all. I cannot keep up. There is a new model every week, a new framework every fortnight, a new platform every month, and I do not know what to ignore and what to learn.
One of them said something that i hear a lot from people. It feels like the ground keeps moving. By the time I sat down to watch the replay, Sundar had already moved it again.
The instinct after a day like today is to add the new releases to the list of things to keep up with. Resist that instinct. The list is the problem, not the solution.
Stop counting features. Start counting layers.
By the end of the morning, Google had shipped a new frontier model (Gemini 3.5 Flash), a new creative model (Omni), a persistent personal agent (Spark), a developer platform for building agents (Antigravity 2.0), an agentic shopping protocol (Universal Cart and the Universal Commerce Protocol underneath it), an agent runtime inside Search, generative UI and mini apps coded inside the search box itself, an agent home screen for Android (Halo), a hardware surface for all of it (Intelligent Eyewear), a provenance layer for AI media (SynthID and C2PA in Chrome and Search), and a pricing reset that halved the entry point for the top tier subscription.
That is not a product launch. That is a stack.
The model. The platform. The runtime. The interface. The hardware. The commerce rails. The trust layer. The pricing weapon. All of it, in one morning.
The companies that have spent three years building one of those layers just discovered the company that built all of them.
This was always the shape
Google has three billion Android devices, two billion Gmail accounts, two billion Chrome users, two billion YouTube monthly logged in users, and the largest search distribution surface in human history. They own Maps. They own the browser most people use to reach the open web. They own a meaningful fraction of the cloud compute layer. They own one of the two relevant mobile operating systems. They own the largest single corpus of human written text and video that has ever been assembled.
You do not assemble that without intent. You assemble it because the announcement is the integration, not the model.
The model was always going to be commoditised. Frontier capability now sits at the lower end of pricing curves that did not exist eighteen months ago. The differentiating layer was always going to be the orchestration of those models against data, distribution, and context that competitors cannot replicate at any price.
Yesterday Google made the strategy visible. The competitive surface is not the model. It is everything underneath, around, and on top of the model.
Six shifts that matter more than the headlines
Most of the coverage will focus on Gemini 3.5 Flash, the new glasses, and the Spark demo. Those are the headlines. They are not the shifts. Six things happened yesterday that will still matter long after the news cycle has forgotten them.
One. The Universal Commerce Protocol. Google shipped more than a shopping cart. They proposed the wire format for agentic commerce. If they make this stick, the shopping cart leaves the merchant’s site for good, and the rails of how AI agents transact with retailers route through Google. Stripe should be thinking about this carefully. Shopify should be thinking about it more carefully. Amazon should already have a team in a room.
Two. Antigravity 2.0. While the consumer headlines went to Spark and the glasses, Google shipped a twelve times faster agent first development platform with native Android support, Workspace integration, and a mobile AI Studio. This is the long tail of agentic infrastructure being absorbed into the platform that owns the models, the runtime, the distribution and the data. The orchestration thesis just got a credible incumbent backer.
Three. The pricing reset. The top tier AI Ultra subscription dropped from 250 dollars to 100 dollars a month, with a five times higher usage limit than AI Pro and a shift from daily prompt limits to a compute used metering model. Google has decided that the consumer agent market is a market they will buy if it cannot be won on product alone. They have the cost curve and the balance sheet to make that decision. Nobody else does, in the same way, at the same time.
One hundred dollars a month, five times the usage of the tier below it. That is the price at which Google has decided the consumer agent market is settled. It is also the price at which most of the consumer AI companies built in the last three years cannot operate. Not because they are badly run. Because the cost curve underneath them is not theirs.
Four. Mini apps inside Search. Search no longer returns answers. It builds them. Generative UI now assembles custom interfaces on the fly, and persistent mini apps, coded by Antigravity and Gemini 3.5 Flash inside the search box, will handle ongoing tasks like fitness tracking, wedding planning, or moving house. The generative UI piece is free for everyone this summer. The persistent mini apps roll out to AI Pro and Ultra subscribers in the US in the coming months.
The implications run deeper than the demos suggested. A meaningful slice of the lightweight SaaS market, the templates, the trackers, the personal dashboards, the single purpose web apps people pay five or ten dollars a month for, just got a free, on demand competitor that lives inside the most used product on the internet. Second, the results page itself starts to disappear. Search stops pointing you somewhere and starts being the destination. The blue link, already wounded, may not survive this. Third, Antigravity is no longer just a developer tool. It is the engine running inside Search. The orchestration platform and the consumer interface are now the same product. That convergence is the most strategically aggressive move on the slate.
Five. Android Halo. A dedicated home for agents on Android, showing what each one is doing in real time. The OS is being rebuilt around agents, not apps. The implication for everyone shipping a consumer app today is that the app icon as the unit of attention has a sunset date.
Six. SynthID and C2PA in Chrome and Search. The least talked about announcement of the day. The trust and provenance layer for AI generated media is being baked into the substrate. If Google makes content credentials a Search and browser default, the economics of synthetic media shift quickly. Provenance becomes a ranking signal. Authenticity becomes a tradeable asset.
The fear underneath the news
If you are building in any of those layers today, the question is not whether your work was good. The question is whether the ground you assumed was open is now occupied. A lot of careful work, by serious people, just got reframed by a keynote.
This is what the calls I had been on were really about. Not the volume of news. The fear underneath the volume. The fear that the thing you have been building is the thing the platform is about to absorb. That you have spent two years learning a framework that is now built into the floor. That you have been told to focus, and now you cannot tell what to focus on.
One of the people on a call yesterday told me she is worried about security. Agents going rogue. The blast radius when one of them gets it wrong inside her business. Another, on a different call, raised something else entirely: the sheer scale of AI slop now flooding every channel he markets through. Two completely different concerns, both real, both surfaced in the same afternoon. The point is not that either of them is wrong. The point is that the platform is now producing problems faster than most operators can name them.
There is no clean answer to that fear. There is an honest one.
The substrate is not the same as the value. Google has assembled the layers. They have not assembled the orchestration of those layers against your customer’s specific reality. They cannot. That part is not on the keynote slide because it is not a Google product. It is the work.
Where the work still is
I work with a marketing compliance business called Flourish. I also co founded Vecto Ventures, where we are building and backing companies that sit in exactly this orchestration layer. The announcements yesterday are deeply relevant to what we are building. None of them threaten the work. All of them strengthen it. The platform now contains primitives that the companies in our portfolio can build on top of, with context (regulatory, brand, audience, jurisdiction, customer relationship) that the platform does not have and will not develop.
The pattern is straightforward. The ground got bigger. The need for orchestration on top of the ground got bigger with it. The companies that succeed over the next decade are the ones that take the platform as a given and build the layer above it that is specific to their customer, their data, their workflow, their judgement.
That is also the answer for the people on my calls yesterday. The instinct when overwhelmed is to try to learn every new tool. That is the wrong instinct. The right instinct is to spend less time on the tools and more time on the human problem the tools are now capable of solving. The capability is no longer the constraint. Attention is. Imagination is. Taste is. Judgement is.
You do not need to keep up with every model release. You need to keep your eye on the problem you are solving for the human in front of you, and let the platform catch up around you.
What Google did not say
There was no serious enterprise story yesterday. No real Cloud reveal. No model that decisively beats Claude or GPT on the hardest reasoning benchmarks. The frontier work that researchers are paid to obsess about was not the centre of gravity. The centre of gravity was consumer, commerce, and distribution.
That is itself a tell. Google has decided the next phase of value capture is not in a smarter model. It is in a wider integration. The model layer is done as the primary competitive surface.
For anyone building in the orchestration layer, that is the signal worth reading. The ground got wider yesterday. The room above it just got more valuable.
The map
The biggest opportunities sit in the disruption, not the disruption itself. Build lifeboats. Offer help. Navigate the waves. Direct people to safety.
Yesterday was the moment the ground showed itself. The water level changed. The response is not defensive. The response is to recognise what is now possible, and to build the unimaginable on top of it.
The lifeboats get built on a different assumption now. The platform underneath you is Google scaled. The room above it, where judgement and context still do the work, is wider than it was on Tuesday morning.
The companies I am building with at Vecto are not reading today as a threat. They are reading it as the largest expansion of the buildable surface in a decade. The model layer is commoditised. The platform layer is consolidated. The orchestration and context layer, where judgement, taste, customer intimacy and human imagination still do the work, is where the next decade gets made.
If you are lost, you are not alone. Everyone is lost. The people who pretend otherwise are mostly selling something.
The work is not to keep up. The work is to focus on what matters: the human in front of you, the problem they actually have, and the layer above the platform where your judgement is still the only thing that does the job.
The ground is now visible. Build above it. Build things that were not possible yesterday.
If this resonated, subscribe. The next anchor piece looks at what the Universal Commerce Protocol actually does to the unit economics of every direct to consumer brand built in the last decade, and where the new defensible positions are.
Craig Hepburn is an AI strategist and Perplexity Fellow. Twenty years building at the frontier of digital, from Microsoft and Nokia to Art Basel and UEFA. Now building at the frontier of agentic intelligence.