jiyu.
← labs

building an IAM agent

identity and access management was built around people. the next generation of systems is built by agents — software that acts on a human’s behalf. the question is not whether an agent can do something, but who it is, who is accountable for it, and what it is allowed to do.

the problem

access without identity is a leak waiting to happen. if we let software act freely, we lose the one thing identity gives us: an answer to “who did this, and on whose authority?”.

deconstruction

an actor is an actor

stop treating “a person” as the definition of identity. a person is one kind of actor; software is another. once identity means who is acting rather than which human is acting, everything downstream — who can do what, who approved it, what changed — works the same way for both.

accountability is the load-bearing idea

every actor is accountable to someone. an agent has an owner — a human who answers for what it does. remove that edge and you get power without responsibility, which is exactly what identity exists to prevent.

acting for someone is different from acting as someone

two abstractions split the field. delegation — an agent acts on behalf of a human, so its actions carry that human’s authority. ownership — even when it acts as itself, an agent is still owned by a human. the difference between “on behalf of” and “as itself” is the whole game.

capability, not power

an agent should not be handed keys and left alone. it should be handed a scope — the smallest surface that lets it do its job, and nothing more. that surface should reflect what its owner can already do, and no more.

tensions

  • autonomy versus control — the agent should act, never outside its scope.
  • fail-open versus fail-closed — when in doubt, deny and ask.

where this lands

these abstractions — actor, accountability, delegation, scope — are what the technical solution is built from. the labs that follow cut through the same problem from two angles.