Synthetic Field Guides

Every high-leverage role deserves a synthetic understudy that can explain the job back to its human mentor.

For the past quarter our lab has been pairing domain experts with lightweight agentic models to capture tacit knowledge. The model listens, drafts instructions, and then the expert aggressively redlines the output. We repeat until the agent can describe the work in clear, verifiable steps. The result is a synthetic field guide—always updated, versioned, and searchable.

The breakthrough isn’t flashy automation; it’s teaching teams how to codify judgment without sanding off nuance. By forcing the model to defend each recommendation with evidence, we surface brittle assumptions in the human process too. It becomes a feedback loop: the agent learns the craft, and the expert sees where the craft needs new instrumentation.

Join the discussion