Let’s talk about why identity is becoming an API.
Imagine opening a new AI product and watching the same old ritual appear in a cleaner costume. Sign in. Connect your email. Sync your calendar. Add your files. Import your contacts. Let the assistant remember you. Let it act for you. Let it become useful by becoming deeply embedded.
It feels new because the interface talks back.
But look closer.
This is an old platform move wearing an AI jacket.
Every major technology cycle has a control point. In the browser era, it was search and traffic. In the mobile era, it was the app store, the operating system, and the notification layer. In the social era, it was identity, graph, feed, and attention.
In the AI agent era, the control point is authority.
Who knows you? Who remembers for you? Who speaks for you? Who books, buys, replies, summarizes, approves, filters, recommends, and negotiates on your behalf?
That is the gameplan.
Big Tech does not only want to give you a smarter assistant. It wants to become the layer through which your decisions pass.
That is why “identity is becoming an API” is not a cute phrase. It is a platform diagnosis.
An API is something other systems call. If identity becomes an API, then your voice, likeness, credentials, memory, preferences, reputation, permissions, and delegated authority become callable resources.
Useful? Absolutely.
Dangerous? Also absolutely.
Because once identity becomes callable, the fight shifts from “who has the best model?” to “who controls the rails?”
This is where the fad gets reskinned.
They used to call it personalization. Then recommendation. Then social graph. Then productivity. Now they call it agents.
Different interface. Same ambition: move closer to the user, collect the context, own the default path, make leaving expensive, and turn the human relationship into platform leverage.
That does not mean every AI agent is bad. That is childish. Agents can remove admin, reduce friction, help doctors, support operators, assist creators, protect people from repetitive work, and make technology feel less like a chore.
But the operator question is not “can an agent help?”
Of course it can.
The question is: who sets the boundary when the agent acts?
If an AI agent sends an email, whose authority is that? If it books a meeting, what scope did it have? If it remembers your preferences, who owns that memory? If it uses your voice, where is consent recorded? If it recommends what to buy, read, believe, or ignore, whose incentives shape the suggestion?
That is the unglamorous part most demos skip.
Capability is exciting. Authority is expensive. So weak products show capability and hide authority.
This is the pattern to watch in every AI launch: how much context does the tool request, how portable is that context, how easy is it to revoke access, how visible are the logs, and who benefits when the user becomes dependent?
When a platform says, “bring us your work,” ask what happens when your work cannot leave.
When it says, “let us remember you,” ask whether memory is a feature or a moat.
When it says, “let our agent act for you,” ask whether you are delegating authority or renting back your own agency through a platform gate.
This is how old fads return.
The CRM becomes the memory layer. The browser becomes the agent shell. The phone becomes the biometric authority device. The productivity suite becomes the operating system for work decisions. The cloud drive becomes the private knowledge base. The assistant becomes the new interface to everything.
None of this is accidental.
It is the map.
If we don’t act, we will start seeing companies hand over operational authority to tools they do not really understand because the demo felt magical.
If we don’t act, we will start seeing users locked into assistants that know them deeply but cannot be exported, audited, or cleanly replaced.
If we don’t act, we will start seeing “personal AI” become the friendliest name ever given to platform dependency.
So the builder’s standard has to be sharper.
Memory must be portable where possible. Permissions must be scoped. Actions must be logged. Consent must be explicit. Provenance must travel with synthetic media. Agents must not exceed the authority they were given. And users must have an exit path that does not require abandoning the digital self they built.
That is the difference between an assistant and a platform trap.
This is why the RADAR lens matters. A SIGNAL tells you what happened today. A THESIS tells you what future is worth defending. RADAR is the uncomfortable middle: it asks who benefits if this pattern keeps scaling.
And the answer is rarely hidden. Follow the defaults. Follow the data flow. Follow the switching cost. Follow the place where the user stops choosing and starts being routed.
That is where the gameplan usually reveals itself.
A useful question is: if this tool wins, what becomes harder for the user to move? Their files? Their memory? Their agent history? Their workflows? Their customer context? Their creative archive? Their trust graph?
Whatever becomes hard to move is probably the real asset.
The next platform war will not look like the last one. It will not only be fought over search bars, app stores, or social feeds. It will be fought over the decision layer: what you see, what you trust, what your agent does, what memory follows you, and which system gets to stand between your intent and the world.
So when the next shiny AI product arrives, do not ask only whether it is impressive.
Ask what it is trying to own.
You’ve got this — if you learn to see the platform move behind the friendly interface.