Production AI agents checklist

Before go-live

Named workflow owner, system of record, "done" definition, and risk class (who must approve).

Schemas for inputs/outputs, versioned rules, tool allow-lists, human queue UX.

Eval set on real cases, redaction policy, rollback path.

After go-live

Watch override rate and silent-fail rate. Silent fails are worse than loud queues.

Change control for prompts, models, and rules—same discipline as application deploys.

People and ownership

Name the business owner, the engineering owner, and the on-call rotation for the workflow. Unowned agents rot.

Train operators on the queue before go-live. Surprises in week one are usually training and UX gaps, not model gaps.

After the first month

Review override reasons weekly. Promote repeated overrides into rules. Retire tools nobody uses. Shrink context that never improves accuracy.

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.