Permissions for AI agents

Map to roles

Define agent identities per workflow with scopes mirroring operator roles—or narrower.

Separate read paths from write paths. Summarizers do not need payment APIs.

Break-glass

Emergency elevated access should be short-lived, logged, and dual-controlled—not a permanent superuser service account.

Identity per workflow

Separate service identities for read-only research agents and write agents. Do not share one bot identity across products.

Rotate credentials on a schedule and on owner change.

Access reviews

Include agent identities in quarterly access reviews next to human accounts. Forgotten bot roles accumulate power.

In practice

Map the workflow on a whiteboard before you open a framework: inputs, systems of record, humans, and irreversible writes. If that map is fuzzy, the agent will encode the fuzz.

Pick ten to fifty real historical cases as an eval set. Include the ugly ones. Run the agent offline against them until critical fields and hard rules are acceptable. Only then connect write tools.

Ship with a pause switch, a human queue, and a weekly review of override reasons. Promote repeated overrides into rules. That loop is how production systems improve—not another prompt brainstorm.

Common failure modes

  • Treating a demo on clean samples as readiness for production volume.
  • One shared service account with broad write access across systems.
  • No owner for the exception queue, so failures pile up as noise.
  • Changing prompts and models without regression gates on real cases.
  • Measuring only model latency or thumbs-up, not completed-case cost and audit completeness.

What good looks like after ninety days

The first workflow is boring: stable override rate, known failure modes, operators who trust the queue. Config changes go through review. Traces answer "what happened to this case?" without archaeology.

At that point you can add a second document type or a second agent role. Expanding before the first path is boring is how programs stall with five half-built pilots.

Platform engineering

常见问题

Service account per agent or per company?

Per workflow/agent class at minimum. One global bot identity is an incident waiting.

How often to review scopes?

With every new tool and on a scheduled access review—like human accounts.