Compliance-oriented AI agents

Right shape

Map controls to checkable assertions, gather evidence from systems, flag deviations to owners.

Humans accept risk or remediate with reasons. Models do not sign the audit.

Wrong shape

A chatbot that paraphrases policy PDFs as if that were continuous control monitoring.

Auto-closing findings because the narrative "looked fine."

Control libraries

Each control needs id, owner, frequency, evidence source, and pass/fail criteria. Free-form policy prose is not enough to automate.

Start with ten well-defined controls, not a hundred vague ones.

Audit use

Export open findings, closed findings with reasons, and evaluation history for a date range. If you cannot export that, the system is not audit-ready.

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.