Give a capable model a vague task — "add user authentication" — and watch one of two things happen. It overreaches: two hours later there are twelve files, eight hundred lines, and nothing works end to end. Or it under-finishes: it declares victory when the code "looks done," before anything was actually run.
These are two faces of the same problem — the agent has no externalized plan and no honest definition of done, so it optimizes for looking productive instead of being finished.
Planning is the cure, and it is not one feature but four, each attacking a different part of the problem.
This week is about restraint, not power. Every tool so far made the agent more capable; this week makes it more disciplined — which, counterintuitively, makes it far more useful.
The walkthrough
Each part adds one capability to the same growing codebase, and each is a runnable checkpoint.
- 01The todo listOne task in progress, always — externalized so it cannot drift.
- 02Plan modeRead-only exploration that proposes but cannot touch your code until you approve.
- 03Feature listsThe harness, not the model, owns what "done" means — verification-gated state.
- 04Loop engineeringWhen to keep going, when to stop, and how to tell the difference.
- 05Maker/checkerThe worker never grades its own homework.
After this week you can
- Separate planning from execution with an approval gate
- Stop scope creep with externalized task state
- Define "done" in the harness rather than trusting the model