I Rarely Use Planning Mode Anymore - Context Windows Are Big Enough
I Rarely Use Planning Mode Anymore
Planning mode was essential when context windows were 8K tokens. You needed the model to think through its approach before writing code because it could not hold the entire problem in context at once. Now with 200K token windows, the model can see everything and figure out the approach as it goes.
When Planning Mode Made Sense
With small context windows, the model had to be strategic about what to focus on. Planning mode forced it to:
- Break the problem into steps before starting
- Identify which files to read first
- Decide on an approach before committing tokens to implementation
This was necessary because a wrong start could waste most of your context window, leaving no room to course-correct.
Why It Is Less Necessary Now
With a 200K context window, the model can:
- Read every relevant file in the codebase
- Hold the entire problem context simultaneously
- Self-correct mid-implementation without running out of room
- Try an approach, realize it is wrong, and pivot - all within one context
The model is effectively doing planning internally. Asking it to explicitly plan first just adds a round trip that produces a plan you then have to confirm before work begins.
When I Still Use It
Planning mode is still useful for:
- Multi-day tasks that span multiple sessions and need a documented plan to maintain continuity
- Team coordination where multiple agents need to agree on an approach before starting
- High-stakes changes where the cost of a wrong approach is very high (database migrations, production deployments)
For everyday coding tasks - bug fixes, feature implementation, refactoring - I skip planning mode entirely and let the model work through the problem naturally.
The context window is the plan.
Fazm is an open source macOS AI agent. Open source on GitHub.