Controlling AI agent cost

Cost stack

Inference, tool infrastructure, engineering, and operator time on exceptions.

A cheap model that doubles overrides can cost more than a pricier model with cleaner fields.

Levers

Cache and stage: do not re-parse unchanged documents.

Route hard cases to stronger models only when rules say so.

Invest in schemas and rules that prevent expensive rework.

Unit economics

Track cost per completed case including human minutes. Token dashboards alone mislead.

A 20% cheaper model that doubles rework is not cheaper.

Engineering cost

Most of year-one cost is design, integration, and ops tooling—not inference. Budget for that or the pilot never graduates.

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.