Agent tutorials and guides

Choose a tutorial to learn with a sample project, or follow a guide to apply an agent workflow to your own codebase in VS Code.

Choose where to start

If you're new to both VS Code and AI, first install the editor and open a workspace. If you already know VS Code, use one of these sample projects to learn the agent workflow:

Already use agents? Skip the sample projects and follow the experienced-agent fast track. If you want AI help without delegating changes, choose a lighter-weight AI feature.

Or jump to a task:

• Work on a project.
• Test and validate.
• Customize and coordinate agents.
• Improve results and recover.

Experienced-agent fast track

Use your own repository and follow the harness, interface, and prerequisites specified by the guide you choose. Keep its recommended session settings for the first task. A guide's tool names or customization files might need changes before you can use them with another harness.

1. Open your repository as a workspace so that files, source control, terminals, tests, and agents share the same project context.

2. Map your existing workflow to VS Code sessions. Check that your account and environment meet the chosen guide's harness setup requirements.

3. Start with a bounded task in your own repository. Explore the codebase without making changes, or add a feature with your existing development environment and tests.

4. Review and validate the code edits before you integrate them.

After one real-project task, adapt agents to your project and compare quality, reliability, and AI credit usage. If you move between supported applications, learn how to view sessions from other applications.

On a managed device, your organization might control which agents and providers are available. Developers can review the agent availability troubleshooting steps, and administrators can manage AI settings. For keyboard, screen reader, and low-vision workflows, use the Accessible View and other accessibility features.

Match a guide to your setup

Before starting a guide, check:

• Project: Does it create a sample project, or require your own repository with working build and test commands?

• Harness: Does it use your selected harness's tools and customization formats? Choosing the same language model doesn't make those formats interchangeable.

• Interface: Which steps use the Chat view or the Agents window?

Both interfaces run agents. The Chat view emphasizes collaboration over code in the current workspace; the Agents window emphasizes multi-project orchestration and higher-level task assignment. Follow the guide's chosen interface for its walkthrough. See Run agents to compare approaches.

Work on a project

• Explore a codebase: trace a behavior without editing files and verify explanations against source references. Use your own repository with Copilot in the Chat view.

• Add a feature: research, plan, implement, and verify a bounded change. Use your own repository with a working development environment and tests, and Copilot in the Chat view.

• Fix an API bug: reproduce a pagination bug, add a regression test, and review the fix. Sample project. Requires Node.js 22 or later and Git.

• Refactor safely: change code structure in reviewable steps while preserving behavior and checking existing callers. Use your own repository with Copilot in the Chat view.

• Work with Jupyter notebooks: create or edit a notebook, run cells, and review analysis results. Use your own data. Requires the Jupyter extension and a notebook kernel.

Test and validate

• Add tests to existing code: generate tests for a function or module, run them, and review assertions, failures, and coverage. Use your own project with Copilot in the Chat view.

• Validate a web app with browser tools: build a calculator app and check it against observable acceptance criteria. Sample web app.

• Set up test-driven development: create custom agents and instructions for a repeatable test-first workflow. Use your own repository with working tests and the Local harness in the Chat view.

Customize and coordinate agents

• Adapt agents to your project: share project instructions with your team, then add skills or specialized agents where needed. Use your own repository.

• Set up a context engineering workflow: curate project context and connect planning to implementation with instructions, custom agents, and prompt files. Use your own repository and the Local harness in the Chat view.

• Run independent tasks in parallel: work in separate Git worktrees, review each result, then integrate and retest the changes. Use your own Git repository with Copilot in the Agents window, after completing a single agent task.

Improve results and recover

• Get an agent back on track: choose whether to steer a request, recover unwanted changes, or reset conversation context.

• Reduce AI credit usage: compare model costs and results, focus context, and scope tools to reduce unnecessary usage.

• Find prompt examples: adapt prompts for exploring code, building features, debugging, and testing.

Background and help

• How agents work: understand the agent loop and the role of models, context, and tools.

• Best practices for using AI: scope tasks, provide relevant context, and verify results.

• Troubleshoot AI features: diagnose setup, sign-in, connection, and other technical issues.