Meta has just published something more important than another model benchmark: an AI architecture in which the durable part of the system lives outside the model.

Meta has just published something more important than another model benchmark: an AI architecture in which the durable part of the system lives outside the model.

Meta calls it an "Organizational Second Brain."

Institutional knowledge is stored in structured, auditable files. Knowledge is separated from reasoning procedures. Expert corrections are compiled into minimal changes, tested through targeted replay and regression, reviewed by humans, and preserved as version-controlled, reversible state.

That is a real architectural shift.

It makes the model more replaceable in principle, because the accumulated knowledge is no longer trapped inside one set of weights.

But this is where the harder problem begins.

Preserving knowledge is not the same as preserving identity.

A new model may read the same files, follow the same procedures and pass the same tests. Yet several questions remain:

Who owns the authoritative state? Who may approve its next version? What exactly is inherited? Do previous permissions remain valid? Is this a continuation, a successor or a fork? What independent evidence proves the transition?

Meta's work concerns organizational memory, not persistent personal identity. It demonstrates a necessary building block, not a complete continuity architecture.

This distinction matters because personal AI may eventually outlive specific models, providers, devices and perhaps companies. At that point, the central question will not be whether the next model can access yesterday's files.

It will be whether it has the right to say:

"I am the same continuing system."

The industry is finally moving durable state outside the model.

Now it must decide who holds that state, who governs its transitions and what continuity actually means.