The TodoWrite Tool: When It Actually Helps

TodoWrite is a tool the agent can call to maintain an explicit task list during a session. It writes the list, marks items in-progress, marks them complete, and re-reads it as it goes. The user sees the list update in real time, which gives the session structure a normal back-and-forth chat doesn't have.

When it helps

1. 1Multi-step work where the agent is genuinely tracking dependencies
2. 2Tasks complex enough that you want to see what the agent has and hasn't done
3. 3Long sessions where re-reading 'what's left' would otherwise burn context
4. 4Hand-offs — you can stop and resume because the list is the state
5. 5Branching work where one decision opens or closes other items

When it's overhead

• Single-step tasks ('fix this typo') — the list is performance theater
• Tasks short enough that the trace itself is enough
• Conversational debugging where structure isn't the problem
• Anywhere the list ends up out of date faster than the agent updates it

Apply: a quality test

1. 1For your next 5 sessions, note when the agent uses TodoWrite
2. 2Mark each as 'helped me track,' 'pure ceremony,' or 'genuinely structured the work'
3. 3Tune your prompts to ask for it on real work and discourage it on simple tasks
4. 4After a week, you'll know your own threshold

The big idea: TodoWrite is for genuinely multi-step work. Skip it on small tasks; use it as a debugging surface on big ones.