Section 1 of 1
How to Govern AI Workflow Releases
Agent BOM is for teams that want to scale AI workflows without losing control. It tracks what an agent is made of, whether the required dependencies are documented, and whether the release record is strong enough for production.
Register the agent
Create the named workflow or automation as a governed entity inside the workspace. Assign an owner so accountability is visible from the start.
Create the version
Add a version label, summary, source reference, deployment target, and runtime metadata such as runtime owner, eval suite, rollback ref, and change ticket.
Build the dependency map
Add model, prompt, tool, connector, dataset, policy, or secret entries. Mark which dependencies are required and record purpose, owner, and control ref.
Open the release gate
Attach an approval to the version. Use checklist items to confirm eval coverage, rollback readiness, and policy alignment.
Release only from a clean record
A version should move to RELEASED only when the BOM is documented, approvals are complete, and graph warnings are acceptably low.
Runtime Owner
Who is accountable when the workflow breaks, drifts, or needs rollback.
Eval Suite
Which test pack or regression set proves the workflow still behaves acceptably.
Rollback Reference
How to recover safely if the release causes harm or degrades quality.
Control Ref
Which policy or control framework governs the dependency or release.
Treat an agent version like a software release record. If you would not deploy code without owner, rollback, and review, do not deploy a workflow without those same control signals.