Study notes · 3.3% of the exam

3.4 Determine when to use plan mode vs direct execution

Choose plan mode for large, multi-file, architectural or ambiguous work and direct execution for small well-scoped changes, and use the Explore subagent to keep verbose discovery out of the main context.

Key points

1
Plan mode has Claude research and propose changes without making them: it reads files and explores but edits stay blocked until you approve the plan. Enter with Shift+Tab until the status bar shows plan mode, prefix a prompt with /plan, or start with claude --permission-mode plan; defaultMode: "plan" in settings makes it the default.
2
Approving the plan exits plan mode and switches to an execution mode so Claude starts editing; Ctrl+G opens the proposed plan in your editor before Claude proceeds. This is the natural way to combine planning for investigation with direct execution for implementation.
3
Use plan mode for large-scale changes, multiple valid approaches, architectural decisions and multi-file modifications: monolith-to-microservices restructuring, a library migration across 45+ files, choosing between integration approaches with different infrastructure requirements, or unfamiliar code.
4
Use direct execution for simple, well-scoped changes: a single-file bug fix with a clear stack trace, adding a date-validation conditional, a rename, a log line, a typo. If you could describe the diff in one sentence, skip the plan.
5
Plan mode adds a full exploration-and-proposal turn, so applying it to everything hurts throughput; the exam's distractor is starting direct and "switching to plan mode only if complexity appears" when the complexity is already stated in the requirements.
6
Plan mode enables safe exploration and design before committing to changes, preventing costly rework when dependencies are discovered late. Detailed upfront instructions in direct execution assume you already know the right structure.
7
The Explore subagent is a built-in read-only subagent for discovery. It reads files in its own context window and returns a summary, so verbose discovery phases do not exhaust the main conversation during multi-phase tasks. Prompt it with "use a subagent to investigate ...".
8
Context is the constraint: performance degrades as the window fills with file contents. /compact and larger models treat the symptom; delegating discovery to a subagent or scoping investigations narrowly treats the cause.
9
Best multi-phase pattern: plan mode plus Explore subagent for discovery, human review of the plan, then direct execution for implementation in the same session, followed by verification (tests, build) before commit.

Read the source

Test yourself on 3.4 Determine when to use plan mode vs direct execution

Ten questions, with the answer and explanation after each one.