CC - Agent-mode

CC - Agent-mode

In the previous section, you built a plan and now you are probably looking at a screen like this:

image-20260428-161103.png

While you could let auto mode take it from here, you would miss out on some explanation. Clear this menu by hitting [escape]

✏️ Press the escape key to clear the prompt

[escape]

Plan mode cannot change any files. To start implementation, Claude Code needs to run in either normal agent mode or auto-mode. While in the normal mode, Claude will ask permission for almost everything. You have the option to deny, approve, or approve similar actions forever. Normal mode will give you maximum control and safety over what changes Claude Code can make, but its also pretty tedious.

What is auto-mode?

By default, Claude needs to ask permission to do anything. You don’t want it deleting a database or sending messages to your ex, so the normal mode allows you to review every action before it runs.

Auto-mode was created to find a balance between safety and being pestered with questions. The Claude team has built a review system that auto-approves a wide range of common actions that you probably want, like writing code and reviewing docs, so you don’t have to monitor the whole implementation process. You will still be asked for permission if Claude tries to perform more unusual or risky actions.

Auto-mode should keep your databases and dignity safe, but there are no guarantees.

If you don’t kick off auto-mode at the end of a plan session, you can switch to it at any time with [shift]+[tab].

image-20260501-170452.png

Implementation phase

Now we're ready to actually build the thing. After reviewing the plan, you could have selected: “Yes, and use auto mode” to start the implementation. Claude Code would then be in full agentic mode with permissions already approved for many common actions. We can also manually switch into auto mode and tell Claude to implement the plan it created.

Pro tip: This is a good time to switch to a cheaper model.

If you used Claude Opus for planning, you can switch to the cheaper Claude Sonnet model for the implementation. The planning phase involves a lot of reasoning and research and Opus made a detailed plan. Sonnet can follow the plan.

✏️ Run the command and select Sonnet

/model

Your plan is ready. You’ve picked the correct model. Switch to auto-mode and tell the agent to start:

✏️ Press [shift]+[tab] until auto-mode is enabled

[shift]+[tab]

✏️ Run this prompt

Implement it

Once you kick off the implementation phase, you'll see Claude do one or more of the following:

  1. Create a to-do list to track its own progress through the plan

  2. Create and edit files — watch them appear and change in VS Code in real time; you can accept or reject individual changes, or approve all at once

  3. Ask for approval before running commands like tests or starting a server — these show up as inline permission prompts in the terminal. If you are in “auto-mode,” you likely won’t see any approval prompts.

  4. Run your tests directly in the terminal and show you the output, iterating if anything fails

  5. Spin up a local server so you can preview the app in the browser

Note: Your experience here may vary depending on what you asked for.

By the end of it, we should have a new "Created by Bitovi" section. Pretty cool.

image-20260428-163945.png

Next Steps

CC - Context Management