Guardrails for AI agents

Layers

Input validation, output schema checks, tool allow-lists, policy engines, human gates, rate limits.

Put the irreversible actions behind the strongest gates.

Testing

Include adversarial and messy inputs in evals. Guardrails that only see happy paths fail open.

Defense in depth

Schema validation, policy engine, tool allow-list, rate limits, and human gates stack. One layer will fail open eventually.

Put the strongest gates on irreversible actions.

Owning changes

Guardrail config is production config. Review it, version it, and test it in CI with adversarial cases.

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.