What you will learn
Routers, Plan validation, Reflection, Adaptive steps, Budget accounting.
Engineering the capability
A planner proposes actions and dependencies. The runtime must check whether each action is allowed, has valid inputs, and fits the remaining budget. A plan can become stale after a tool result, so validate again when executing rather than approving the entire plan as an unrestricted script.
Routing selects a workflow based on task properties. Begin with clear deterministic rules where possible and evaluate ambiguous cases. Reflection is a revision step using feedback; without new evidence or a concrete rubric, a model can confidently replace a correct answer with a worse one.
Bound reflection and replanning. Persist the reason for a change and the evidence that triggered it. Distinguish a recoverable tool error from a request requiring human clarification. Repeated replanning should not conceal that the system lacks enough information to proceed.
Worked case
A draft cites a policy that does not mention the requested exception. A useful reflection step checks claim support and requests a targeted search. “Make the answer better” without a rubric may merely add persuasive wording. Evaluate the revised answer against evidence, not its length.
Put it into practice
Continue with the next lab: compare fixed, routed, and reflective flows. Build the artifact, record the failure cases, and explain the tradeoff before moving on.