Plan Mode
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
Plan Mode is a workflow configuration in Claude Code in which the agent pauses to produce an explicit planning step before taking actions, allowing users to review and approve the proposed approach.
Superseded by Auto Mode
The speaker in Reflecting on a year of Claude Code describes having moved away from Plan Mode entirely in favor of Auto Mode: "What they used to be is plan mode. I don't use that anymore. I use auto mode" (10:49, 10:53).
Two distinct arguments are offered for why Plan Mode has become unnecessary:
Model capability improvement: Newer Claude models (starting with version 4.6) no longer require a discrete planning step as part of their reasoning process, making the mode architecturally redundant — "newer models they don't actually need like a planning step anymore" (11:39).
Safety argument against granular permission prompts: The speaker argues that Plan Mode's implicit promise of user oversight through per-action permission prompts is actually counterproductive. Because users will approve the vast majority of prompts, they become desensitized and stop paying close attention — making auto mode safer in practice: "auto mode is more safe than reading every single permission prompt" (13:27). The logic is that limiting interruptions to genuinely important decisions preserves user attention for cases that matter.