Now that Claude has access to your files, context, and memory, it can actually do things — not just answer questions. It can read a file, analyze data, write output, and save it somewhere. That is powerful. And with power comes a question: how do you stay in control?

The answer is plan mode and execution mode.

What Plan Mode Is

In plan mode, Claude describes exactly what it intends to do — before it does anything. It presents a step-by-step plan and waits for your approval. Nothing changes until you say go.

This is especially important when Claude is about to:

• Write to a file
• Edit an existing file
• Perform multiple steps in sequence
• Make decisions that affect your data or your business

You review the plan. If it looks right, you approve it. If something is off, you correct it before any action is taken.

What Execution Mode Is

In execution mode, Claude acts. It reads files, writes output, completes tasks. This is what you want most of the time — but only after you have confirmed the plan looks right (for consequential tasks) or when the task is low-stakes enough that you do not need to review first.

When to Use Each

Situation	Use Plan Mode First?
Claude is about to edit or overwrite a file	Yes — always
Multi-step task (analyze → summarize → save)	Yes — review the plan before it starts
Writing a new draft you will review before using	Optional — up to you
Simple analysis where Claude is only reading, not writing	No — execution is fine
Answering a question or explaining something	No — no files are affected

How to Activate Plan Mode

In Claude Code, you can activate plan mode in two ways:

Option 1: Ask for it explicitly

Before giving Claude a complex task, type: “Before you do anything, give me a plan for how you would approach this.” Claude will describe its intended steps. You review and approve before it proceeds.

Option 2: Use the Plan Mode toggle

Claude Code has a built-in plan mode. When enabled, Claude will always present a plan before executing. You can find this in the Claude Code interface — look for a “Plan” or “Auto-approve” toggle depending on your version.

Example: Plan Mode in Action

Let’s say you want Claude to analyze this month’s PPC bulk file and flag high-ACoS keywords.

What you type:

“Look at reports/ppc-bulk-april.csv, identify every keyword with ACoS above 40%, and save the results to analysis/ppc-flags-april.md. Before you do it, show me your plan.”

What Claude says in plan mode:

Here is my plan:

1. Read reports/ppc-bulk-april.csv
2. Filter all rows where the ACoS column exceeds 40%
3. Group results by campaign and ad group
4. Write a summary to analysis/ppc-flags-april.md with flagged keywords, their current ACoS, and suggested bid adjustments

Does this look right? Should I proceed?

You review it. Maybe you want to change step 3 — you want keywords sorted by spend, not grouped by campaign. You tell Claude: “Sort by spend descending instead of grouping by campaign.” Claude updates the plan, you approve, it executes.

This back-and-forth takes 30 seconds and saves you from discovering a problem after Claude has already written a file you now have to re-do.

Exercise

1. Find a CSV or report in your /reports folder (or add one now)
2. In Claude Code, ask: “Look at [filename] and summarize the top 5 insights. Before you do it, tell me your plan.”
3. Review the plan Claude presents
4. Approve it or modify one element before approving
5. Let Claude execute and review the output

Once you are comfortable with plan-then-execute for complex tasks, you will have a reliable workflow that keeps you in control regardless of how sophisticated Claude’s actions become.