> ## Documentation Index > Fetch the complete documentation index at: https://docs.cogniagent.ai/llms.txt > Use this file to discover all available pages before exploring further. # Plan mode and Build mode > Have your AI coworker research and propose an approach before it changes anything — then switch it loose to do the work. Agree the approach before the work starts. It's the cheapest way to avoid a confidently wrong afternoon. ## The two modes **Plan mode** — your coworker can look but not touch. It researches, reads files, works out what's involved, and comes back with a proposed approach. Nothing is changed, nothing is sent. **Build mode** — the full toolset. This is the normal way a coworker works. Switch with `/plan` and `/build`. The current mode shows as a chip under the message box, so you always know which one you're in. The message box with the Plan chip active alongside the effort and model chips ## Why plan first Your coworker is fast, which is exactly the problem on a job where the wrong approach is expensive. Plan mode puts a checkpoint between "I understand the task" and "I've done it". It's also the fastest way to find out you were misunderstood. A plan takes seconds to read; an afternoon's work takes an afternoon to undo. ## When you'd use it * Anything touching live customer data * A job you can't easily undo — sending, deleting, publishing * Work you're going to be judged on * A task you've described badly and know it * The first time you give a coworker an unfamiliar kind of job * Anything expensive, so you're not paying for the wrong thing twice ## How it goes Type `/plan`, then describe the job. Your coworker researches without changing anything. Steps, and usually the assumptions it's making. The assumptions are the interesting part — that's where a misunderstanding shows. "Skip step 3" or "the file is the second one, not the first". Still nothing has happened. `/build`, and it does the work you agreed. ## Good to know **Read the assumptions, not just the steps.** The steps usually look fine. The line that says "I'll treat blank rows as zero" is the one that matters. Plan mode is genuinely read-only, not a promise to behave. It's a safe way to point a coworker at something sensitive. You don't need it for everyday work. Reserve it for jobs that are expensive, irreversible, or badly described. ## Common questions It's the agreed approach, not a script. If reality differs — a file isn't what it expected — your coworker adapts and tells you, rather than following a plan that's stopped making sense. Yes. Switch back at any point, which is useful when a task turns out bigger than expected. A little, and it routinely saves more by preventing work you'd have thrown away. Plan mode checks the approach **before**. [Thorough review](/cowork/chat/thorough-review) checks the result **after**. On genuinely important work, use both. ## Next steps The check at the other end. What pauses for you anyway. How to switch. Following the plan as it runs.