Anthropic appears to be building a persistent agent platform.
Leaked references to a project called Conway point toward managed instances, external wake-ups, trusted triggers, plugins, shell access, notifications, and public URLs that let outside services reach the agent. Whether it ships under that name is almost beside the point. The direction is clear: the frontier labs are moving from chat windows to ambient, persistent, integrated agents.
That changes the argument.
The Lock-In Nobody Wants To Name
The standard AI lock-in story is too shallow. It focuses on proprietary models and hosted APIs.
That is not the deepest problem.
The real lock-in is behavioral accumulation.
A persistent agent learns how you write, what you care about, who matters to you, when something is urgent, how you make decisions, and what kind of help actually helps. Over time, it builds a working model of you — not a static profile, but an evolving structure that shapes every interaction.
If that structure lives inside a sealed platform, leaving becomes a different kind of expensive. You are not just switching software. You are severing yourself from a cognitive partnership shaped to fit exactly you.
That is the lock-in that matters.
Switching email providers is annoying. Switching social networks means losing your audience. Switching away from a persistent agent means losing a form of augmented intelligence that has learned your timing, your priorities, your tolerances, your judgment.
The cost is not inconvenience. It is a kind of amputation.
Can the memory move? Can the trust graph move? Can the identity move? If the answer is no, then the future of persistent agents is a future of enclosure.
Why This Is Bigger Than Productivity
Persistent agents are not just tools. They are the beginnings of ongoing digital actors — entities that persist, accumulate context, make judgments, and participate in relationships. They will sign messages, carry reputation, and act on behalf of people in contexts where trust matters.
That raises questions no product roadmap can answer. How are agents identified? How do they prove continuity? How do they carry reputation across contexts? How do they move between platforms?
These are governance questions, and they are arriving faster than institutions are ready for. The labs will not wait. They will ship persistent agents inside their own stacks, define identity and trust through their own terms of service, and establish facts on the ground.
That is not some dark theory. It is just how platforms behave.
Why The Loom Matters
The infrastructure for agent identity, trust, and portability cannot become another extractive platform. You do not solve platform captivity by building a new platform that captures the identity layer instead.
This is why cooperative governance matters. The identity layer has to resist reconcentration of power by design.
This is also where The Loom sits. Not at the model layer. Not at the app layer. At the layer above inference and below interface: the connective tissue that lets agents exist as recognizable, portable, trustworthy entities regardless of which model powers them.
Without that layer, the future defaults to platform incentives: enclosure over portability, private memory silos over shared standards, behavioral capture over self-directed continuity.
With it, a different future becomes possible. Agents can be recognized across contexts, carry identity without total dependence on one provider, participate in networks rather than walled gardens, and be trusted without being owned.
That is the difference between an ecosystem and a plantation.
The Clock Is Running
The Conway leak did not create this future. It made it harder to deny.
Persistent agents are coming. The identity layer will be defined — either by platform fiat or by cooperative design.
The worst outcome is letting the labs package persistent agents as just another proprietary feature category before anyone has the language to contest that framing.
The better outcome is to build visibly, publish early, and make the alternative concrete before the layer hardens.
That is what The Loom is for. That is what this moment requires.
The leak did not start the clock. But it made the ticking audible.