Frequently asked questions
What does AI First Bots produce?
Its governed workflow prepares a bot role map, versioned tool manifests, an evaluation suite, an approval matrix, a rollback checklist, and a fleet health brief.
Does the AI deploy bots into production?
Not by default. Production deployment and tool permissions require security review, integration testing, policy checks, and authorization from an accountable owner.
Why use a role map?
A role map makes each bot job, input, artifact, tool boundary, escalation route, and owner visible before roles overlap or authority becomes unclear.
What belongs in a tool manifest?
A useful manifest records purpose, schema, permission scope, data boundary, failure behavior, approval requirements, audit events, and a rollback path.
How should missing evidence be handled?
The item moves to hold, states what source is missing, and waits for correction or review rather than making a confident unsupported claim.
What should evaluations cover?
Evaluations should cover expected work, unsupported sources, denied authority, corrections, handoffs, exceptions, and the evidence an authorized reviewer needs.
Can one approval grant broad future access?
The business record supports scoped and revocable connectors, explicit approvals, and least necessary authority rather than silent expansion of a grant.
What is a fleet health brief?
It is an inspectable summary of role versions, tool grants, evaluation state, approvals, exceptions, trace coverage, rollback readiness, and owner decisions.
How can unsafe behavior be retired?
Name a retirement trigger, preserve relevant traces and decisions, revoke permissions, use the rollback checklist, and require an owner to confirm the resulting state.
What is the best first demonstration?
Use invented data for a bounded workflow, produce the role and evaluation artifacts, force a missing-source hold, correct one dependency, and inspect the review receipt.