Two different modes, two different AI models, one clean handover. Here's the whole story.
Claude Code has a safety switch. In Plan mode it can only look and think. In Do mode it can actually change things.
Claude reads your code, asks questions, and writes a plan — but touches nothing.
Claude reads and acts. It edits files, runs tests, installs packages — with your approval.
Claude Code uses a smarter-but-slower model to plan, then a faster-but-cheaper model to execute. The plan file is the bridge.
Everything that happens from "start planning" to "code shipped."
It sounds fancy. It's not. Here's the whole truth.
Walks the land, asks what you need, draws up the blueprints. Doesn't touch a single brick. Smart, slow, charges a lot per hour.
Blueprint = plan.md
The one thing that crosses from planning to building. You (the client) approve it before anything gets built. No blueprint = no building.
= ExitPlanMode + your "yes"
Reads the blueprint, gets to work. Doesn't need to know why the architect made those choices — just needs clear instructions. Fast, efficient.
Edits files, runs tests, ships