AI agent orchestration

What orchestration must own

State transitions, retries, timeouts, idempotency, and who is allowed to run which step.

Correlation ids across tools so support can answer "what happened to case 4821?"

Anti-patterns

A planner agent that invents steps with no allow-list.

No dead-letter queue—failed runs vanish into logs.

Orchestrating in a spreadsheet of zaps with no single case id.

Idempotency

Retries will happen. Tool calls that create tickets or payments must be safe to retry or use idempotency keys.

Without that, a blip becomes a double post and a trust problem.

Visibility

Operators need a case timeline: steps completed, waiting on human, failed tool, next action. Orchestration without a timeline is ops debt.

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.