Agents · Post-Training Engineering

Turn production failures into reproducible decision problems. First determine whether the tool, context, or model needs to change; then connect training data, interactive environments, verifiers, and release acceptance.

Chapters

1. Chapter 25: Agent Post-Training: From Failed Trajectories to Reliable Policies

Related reading

• Tool Learning: tool trajectories and role masks.
• Post-Training Methods: supervision signals, optimization methods, and data generation.
• Agent Evaluation: task outcomes, behavioral constraints, and operating costs.

Back to Agents.