Anthropic just shipped the most useful feature your company will adopt this year, and the most dangerous one. They are the same feature. The firms that understand why will pull away from the ones that pick a side. The interface is disappearing. The AI is moving into the org chart. That shift decides who owns the next decade.
Anthropic's newest release does something quietly radical. It puts Claude inside the place your team already works, the chat, the channels, the documents, and gives it memory of everything your company does. You no longer go to the AI. The AI sits in the room. Andrej Karpathy called it the third paradigm of how we use these models. First a website you visited. Then an app you opened. Now a persistent coworker that lives where the work happens and never logs off. Hold the opportunity and the danger in the same hand. They are the same feature. Any company that sees only one of them gets this wrong.
The Opportunity, Said Plainly
Start with why you will adopt this, because you will. Most of what a company spends its energy on is not the work. It is the talking about the work. The status update. The handoff. The meeting that exists only to align two people. The new joiner who takes three months to learn where anything is. There is a polite phrase for all of this, human middleware, and it is a large share of what your payroll actually buys.
An AI coworker with full context dissolves a lot of it. It remembers what was decided and why. It onboards in seconds because it already knows everything. Point it at a goal and let it run against that goal around the clock. For a services firm this is not a productivity tweak. The delivery layer becomes native. Scope, evidence, handoff, the institutional memory that leaks time and trust on every engagement, becomes something the system holds and acts on. That is real. It is large. Refusing it on principle is how a firm loses a decade. So take it. I am not here to tell you to be a luddite.
The Danger, Said Precisely
Now the part the productivity decks leave out. The moment your AI vendor becomes a coworker, it stops being a model you call. It becomes the place where your work is interpreted, remembered, routed, and eventually done. To make it useful you feed it everything. Every document, every conversation, every process, every decision. In return you get leverage. You also hand the vendor a live map of how your company actually works.
That is not model lock-in. You can swap models in an afternoon. This is context lock-in, and context does not move.
Your operating system now lives in someone else's house, and you rent it back by the token. See where that leads. The vendor sees everything, so it can price with everything it sees. It can step into any adjacent market your data reveals, the way every platform eventually competes with the ecosystem built on top of it. The cost scales with ambition. If the winning move is to point near-infinite tokens at grow my business and loop forever, the firm that can afford the most compute improves the fastest, and the gap compounds. Intelligence stops being a tool you own. It becomes a utility you meter. Utilities concentrate.
This is the real economic story, and it is bigger than any single product. The economy is about to split twice. First, between the companies that become AI native and the ones that do not. That is the productivity divide everyone is watching. Second, and almost nobody is pricing this one, between the companies that still own their judgment and the ones that quietly became tenants in their own operations. The second divide is the one that lasts.
The Clever Part, Which Is the Whole Point
Here is the synthesis. It is the opposite of both the doom take and the hype take. When the work itself becomes rentable, value migrates to the part that cannot be rented.
I have argued a narrower version of this before. As AI makes analysis abundant, the scarce thing becomes conviction, the willingness to commit under an ambiguity no model resolves. This is that argument at the scale of the whole firm. You can rent the execution. You cannot rent the judgment about what to execute, the proprietary context you choose not to surrender, the decision to bet the company on one direction over another. None of that comes from the vendor. It is the only thing that was ever really yours.
When the work itself becomes rentable, value migrates to the part that cannot be rented.
So the move is not to refuse the coworker. It is to rent the work and own the conviction. In practice that means adopting the productivity aggressively while keeping sovereignty over the thin layer that matters. Hold a judgment layer you do not cede. Keep the proprietary data that is your actual moat out of the shared graph, or in a form you control. Keep the ability to leave. Treat the AI as rented labour, brilliant and tireless and not yours, not as the new owner of your operating model. The firms that win the next decade will not be the ones that adopted earliest or resisted longest. They will be the ones that took all of the leverage and gave away none of the conviction.
The Incentives Are Obvious, Not Sinister
Anthropic is not doing anything sinister here. The incentives are simply obvious, and obvious incentives are the kind you plan around, not the kind that should surprise you. So adopt the coworker. Feed it the work. Take the productivity, all of it. Just remember whose name is on the lease. Rent the body. Keep the spine.