Where Influence Actually Accumulates in the AI Era
When Gemini’s big update landed, the reactions surfaced in a familiar pattern. People didn’t just celebrate; they moved. Confidence shifted, curiosity spiked, timelines filled with screenshots and first impressions. Some even announced they were switching models. Whether they stay or drift back isn’t the point. The behavior reveals something simpler: when it comes to AI models, loyalty is light.
Most people open a model the way they open a tab. It’s a tool, not a home. And tools get replaced the moment something sharper appears.
That doesn’t mean models don’t matter. It means influence might not settle where we expect.
The place that becomes hard to leave is rarely the thing that improves fastest. It’s the thing people pass through without thinking. And those places look very old and very familiar: phones, browsers, operating systems, account ecosystems.
AI isn’t rewriting these systems. It’s turning some of their entry points into places that are simply harder to walk away from.
Models Are Easy to Try and Easy to Leave
Every major model still looks roughly the same from the outside. A text box. A history of conversations. A reply streaming line by line. Move from GPT to Claude to Gemini and the experience doesn’t fracture. You simply land somewhere new and keep going.
That sameness lowers the cognitive cost of switching. Better performance becomes a good enough reason to move. Release cycles replace loyalty cycles.
There is a form of friction, but it’s soft. Over time, people learn the "feel" of a model. What phrasing works. How to nudge it. Which shortcuts it seems to understand. Once you switch, some of that tacit knowledge evaporates and must be rebuilt.
But most users don’t accumulate enough of that familiarity to anchor themselves. They use AI the way people use calculators. Open it. Get the result. Close it.
Tools you open on purpose tend to behave like this. Tools you pass through automatically do not.
The Layer That’s Becoming Heavy
If influence doesn’t accumulate at the model part, the next question is obvious: where does it go?
Look at the places people can’t avoid:
- unlocking a phone
- opening a browser
- signing into the same account that runs their email, calendar, storage, and payments
These actions happen dozens or hundreds of times a day. And they happen without conscious decision. No one wakes up and thinks, “I am now choosing Safari.” They just tap.
This difference matters. Habits formed at low consciousness settle deeper. They resist change. And the deeper the habit, the heavier the switching cost.
That cost compounds through three forces:
- Frequency: repeated actions form paths that are hard to break.
- Accumulated data: photo libraries, passwords, documents, settings, histories.
- Cross-service binding: when your ID also manages payments, devices, subscriptions, files.
You don’t leave an ecosystem like that because something else feels slightly better. You leave only when staying becomes impossible.
Models don’t benefit from any of these forces. They don’t own the objects people return to fifty times a day. They don’t store irreplaceable data. They don’t anchor a network of other necessities.
Not yet.
When AI Moves Into the System Instead of Sitting Behind a URL
The clearest shift in the last year wasn’t just model quality. It was deployment.
Small models began running on phones and laptops. System-level AI started showing up in texting, search bars, photos, apps, keyboard suggestions. Instead of “opening AI,” AI appeared wherever the user already was.
Once this happens, the system, not the model, becomes the gatekeeper. The device decides what runs locally, when to call the cloud, which model to invoke, and how the output appears.
In other words: the contact point becomes the orchestrator.
This isn’t settled. The cloud–edge split is still fluid. Open-source stacks are maturing. Regulatory winds remain unpredictable.
But directionally, the leverage seems to be drifting upward in the stack. Toward the operators of the places people already live inside.
Google, Apple, Microsoft: Three Very Different Kinds of Gravity
Each of the major operators sits on a different form of contact point gravity.
Google’s gravity is reach. Search, Maps, YouTube, Chrome, Gmail, Android. The company updates the world through the cloud. Millions feel the difference the same week. Google’s challenge isn’t speed but depth. A fragmented Android landscape and constant regulatory scrutiny keep the foundation uneven.
Apple’s gravity is cohesion. Hardware, OS, and identity sit in a single architecture. AI slides into daily actions without requiring user choice. The boundary is smaller but denser. Long-term value depends on whether Apple turns cohesion into something more ambitious than a polished shell over familiar workflows.
Microsoft’s gravity is replacement cost. Enterprise workflows live inside Office, Teams, OneDrive, and Azure. Moving these systems is expensive and politically fraught. Organizations rarely migrate once fully committed. Microsoft’s challenge is simple: enterprise gravity doesn’t always spill into consumer life.
Three positions, three different types of accumulation.
The Outliers: Meta and Amazon
Meta proved a model can reach massive scale through distribution alone. Llama touched hundreds of millions across WhatsApp and Instagram. But social platforms remain optional. A phone is not.
Amazon showed that presence without necessity doesn’t create lock-in. Alexa lives in millions of homes, yet remains easy to ignore. Bezos once said customers are beautifully dissatisfied. Looking at Alexa today, that dissatisfaction still feels visible.
The difference between “everywhere” and “unavoidable” is the difference between influence and gravity.
Watching What Moves Next
Some forces move fast: reach, visibility, deployment speed. Others take longer: retention, habit formation, data accumulation, system embedding. And some redraw the map entirely: regulation, new interfaces, shifts in what people expect.
Buffett liked to say that the best businesses are the ones people cannot stop using. The same idea echoes here. Influence tends to settle where people return without thinking, not where the newest feature lands first.
Influence will not settle inside a single part. It will distribute. Some places will gain weight. Others will lighten. And the movement between them might be the signal worth watching.
The question isn’t who wins the model race. The question is which positions become harder to walk away from, and which parts begin to matter more than the models running on top of them.
The structure is still shifting. But the early contours point to something simple. And if general-purpose humanoids eventually enter daily life, the same rule applies. The machines people cannot avoid, the ones that end up occupying physical space in homes or workplaces, will matter more than the ones that simply impress on first use.
The places people cannot avoid often outlast the tools they choose on purpose.
