interactive · 12 min

A Plan-Execute Loop

Objective: turn a goal into an ordered plan, execute each step, and collect the step results — the plan-execute pattern.

Mock is the default and free. Switch the Model toggle above to Real (your key) to run this same code on your own OpenAI-compatible model.

MockPlanner is the Foundations MockLLM seam in planner clothing: a deterministic goal -> plan. In real mode, the model returns a comma-separated list of steps that parse_plan splits into the same list shape; empty output yields [] (the mock’s own default for an unknown goal), while a non-conforming chatty reply degrades into a best-effort list — an honest reminder that real output is unreliable and the parser is where you would harden it. Keeping planning and execution separate means a bad plan is inspectable and replaceable without touching the executor.

Best practice: keep planning and execution as separate units — you should be able to inspect or swap the plan without changing how steps run.

Next: Reflection & Its Limits