Hand Off Tasks, Get Results

Agent mode lets AI work autonomously on coding tasks. You describe the outcome, AI plans and executes, you review the results.

Here’s how to use agent mode effectively.

What Agent Mode Does

Agent mode:

1. Takes a task description
2. Plans the approach
3. Reads necessary files
4. Makes changes
5. Tests the changes
6. Reports results

You stay in control but don’t do the mechanical work.

Good Agent Tasks

Add a feature: “Add pagination to the user list API”

Write tests: “Write unit tests for the authentication module”

Refactor: “Extract the email sending logic into a separate service”

Bug fixes: “Fix the off-by-one error in the pagination calculation”

Documentation: “Generate API documentation for all controllers”

Task Description Quality

Vague (poor results): “Make this better”

Clear (good results): “Refactor this function to:

• Extract the database query into a separate method
• Add error handling for null results
• Add a unit test for the edge case where user.age is undefined”

The clearer the task, the better the results.

Watching Agent Work

Monitor as the agent executes:

• See which files it reads
• Watch changes being made
• See tests being run
• Catch issues early

You can pause or stop if something goes wrong.

Approval Workflows

Configure what needs approval:

High autonomy:

• Agent makes changes automatically
• You review final results

Controlled autonomy:

• Agent plans, you approve the plan
• Agent executes, you review results

Step-by-step:

• Approve each significant change
• Maximum control

Choose based on task sensitivity.

Iteration

First result not perfect? Iterate:

Agent result: PR with new feature You: “The tests pass but the error messages aren’t user-friendly. Update them to match our existing error style.” Agent: Makes targeted improvements

Refinement is expected.

Multi-File Tasks

Agents excel at multi-file changes:

“Rename the ‘getUserById’ function to ‘findUser’ across the entire codebase”

Agent:

• Finds all usages
• Updates each file
• Maintains consistency
• Runs tests to verify

What would take you an hour takes minutes.

Testing with Agents

Agents can run tests:

“Run the test suite and fix any failures you introduced”

Agent:

• Runs tests
• Identifies failures
• Analyzes cause
• Makes fixes
• Re-runs to verify

Automated quality assurance.

When Not to Use Agent Mode

Learning situations: If you need to understand, use chat instead.

Critical security code: Don’t delegate security-sensitive code without review.

Unclear requirements: If you don’t know what you want, use chat to explore first.

Novel problems: New ground may need human creativity.

Agent + Human Review

Always review agent output:

• Read the changes
• Understand what changed
• Run additional tests if needed
• Approve or request modifications

Agent accelerates work. Human ensures quality.

The Agent Mode Checklist

For effective agent usage:

• Clear task description
• Defined expected outcome
• Appropriate autonomy level
• Monitor execution
• Review all changes
• Iterate as needed

Delegate the work. Keep the control.

Use agent mode in AI IDE →