From agent sprawl to an accountable fleet
Somewhere in your organization: a hackathon support bot, three almost-identical code reviewers, and an agent with database access nobody remembers granting. Foundry puts every Worker in one registry—with an owner, a version, an eval status, and an authority boundary.
The sprawl audit
How many agents are running? What is each allowed to do? When one team improves an agent, who else benefits? When an agent misbehaves, how do you find out? Foundry turns those questions from archaeology into inventory.
From sprawl to fleet
The same merge-gate discipline engineers already trust, applied to AI behavior.
Every Worker, Blueprint, Card, version, owner, authority boundary, and eval status in one place.
Build shared judgment once as a versioned Capability Card, then let every eligible Worker inherit it.
Apply risk-proportional eval and review policies before a behavior change can reach production.
The propagation advantage
One team improves a security-review Card and records measured uplift. Every other Worker mounting that Card receives the proposal through its own eval gate and approval policy. Fleet capability compounds without bypassing local control.
Two on-ramps, one fleet
No big-bang rewrite. Most organizations use both paths in the same week.
Deploy versioned Blueprints with eval suites and maintainers—not abandoned template snapshots.
Register existing agents with thin adapters so day one produces the inventory.
Add evals to the Workers that matter most first, then converge duplicated capabilities during normal work.
Field notes on registries, authority boundaries, eval policy, propagation, and adoption.
No published posts match this feed yet.
Next step
Every Worker listed, gated, owned, and able to compound what the organization learns.