Production checklist

AI Agent Production Readiness

An agent is not production-ready because it completed a demo. Review these six systems before it receives real business context, permissions, or customers.

1. Workflow

✓The job has a clear trigger, inputs, outputs, and completion condition.
✓Success and failure are measurable.
✓Ambiguous decisions have explicit escalation rules.
✓The workflow is worth automating frequently enough to justify the complexity.

2. Context

✓Authoritative data sources are defined.
✓Freshness requirements are explicit.
✓Conflicting sources have a resolution rule.
✓The agent cannot silently substitute low-quality context.

3. Tool contracts

✓Every write action has a verifiable success receipt.
✓Retries, timeouts, and rate limits are handled.
✓The agent knows when to stop instead of looping indefinitely.
✓Tool failures are visible to operators.

4. Permissions

✓The agent uses least-privilege access.
✓Read and write permissions are separated where practical.
✓Consequential actions require approval.
✓Credentials can be revoked without taking down unrelated systems.

5. Observability

✓Runs are logged with enough evidence to reconstruct what happened.
✓Accuracy, completion, latency, cost, exceptions, and human corrections are measured.
✓Business outcomes are tracked separately from task completion.
✓There is a recurring evaluation set for regression testing.

6. Exception handling

✓The agent can identify when information is incomplete or contradictory.
✓It can stop safely rather than forcing an answer.
✓Escalations include the evidence a human needs to decide.
✓High-risk edge cases have a documented fallback path.

Decision rule

If you cannot answer one of these sections confidently, reduce the agent’s permissions or scope before increasing autonomy. Reliable automation compounds. Unobserved automation compounds mistakes.

If you already have a real workflow, move from checklist to evidence.

See a sample preflight report →Request AgentOps pilot →Agency partner pilot →