[How Coding Agents Work][13] Planning

How Coding Agents Work · Module 13 of 29: Planning Part II — The Six Layers Control layer. Whether an explicit plan helps, and how Codex represents one. The plan and spec-plan tools, how plans are surfaced and updated, and the benchmark evidence on structured pipelines versus open-ended loops. After this module you can decide whether your task class benefits from an explicit plan, and design a plan representation the model will actually maintain. Lessons in this module: 1. Why plans exist: drift, coherence, and the user's need to see intent 2. The plan tool: representing and updating a todo list, and the rules about when not to plan 3. Plan mode as a behavioural preset: a mode that forbids mutating actions until explicitly exited 4. Modes versus tools: why a read-only session mode and a checklist tool are different mechanisms 5. When structure beats autonomy: the localization–repair–validation evidence Prerequisites: module 12. About the course: Built from the OpenAI Codex harness at commit 8d01cd42fa.