> AI agents: For documentation discovery and navigation, see [llms.txt](/llms.txt).


# Build with AI in VS Code

Use AI in Visual Studio Code to understand unfamiliar code, fix bugs, and build features. For a task that spans files and tools, an agent can find relevant code, make changes, and run checks without you directing each search, edit, and test run. You describe the outcome, review the changes, and verify the result.

Work with an agent in the same workspace as your editor, terminal, tests, and debugger. You can inspect its changes and investigate failures without moving code and command output to a separate chat application. For a question or focused edit, use chat, inline chat, or suggestions without delegating an entire task.

Choose from multiple [agent harnesses](/docs/agents/concepts/agent-harnesses.md), including Copilot, Anthropic Claude, and OpenAI Codex, or delegate independent tasks to [cloud agents](/docs/agents/run/agent-harnesses.md#start-a-cloud-session). VS Code provides a shared chat, session-management, and change-review experience while each harness provides its own tools and workflows. You can also bring your own model API key and extend agents with tools and plugins to fit your team's requirements.

When you use the Copilot harness, you get a consistent agent experience across VS Code, GitHub Copilot CLI, and the GitHub Copilot app. These experiences share the Copilot agent runtime, so you can reuse supported project guidance, such as Agent Skills, across them.

<div class="docs-action" data-show-in-doc="true" data-show-in-sidebar="false" title="Try your first agent task">
Build and validate a small app in the Chat view, then review the result.

* [Start the agents quickstart](/docs/agents/quickstart.md)

</div>

If you're new to both VS Code and AI, first [install the editor and open a workspace](/docs/getstarted/overview.md). If you already use agents, skip the sample project and follow the [experienced-agent fast track](/docs/agents/guides/overview.md#experienced-agent-fast-track). For AI help without delegating a task, [choose between suggestions, focused edits, questions, and agents](/docs/agents/reference/ai-features-cheat-sheet.md#choose-a-feature).

The quickstart uses the [GitHub Copilot harness](/docs/agents/run/agent-harnesses.md#use-the-copilot-harness). See the [Copilot setup guide](/docs/setup/copilot.md) for account, usage, and data-handling requirements. If provider, sign-in, or execution location affects your setup, [understand the session controls](/docs/agents/run/agent-harnesses.md#understand-the-session-controls) and [choose an agent harness](/docs/agents/run/agent-harnesses.md#choose-a-session-target).

## What you can do with AI

Choose a workflow based on the task, from help with a single line to a coordinated change across your project.

| Your goal | How AI helps |
|-----------|--------------|
| Implement a feature, fix a bug, or refactor across files | [Work with an agent](#how-you-work-with-an-agent) to find relevant code, make changes, and run tests. |
| Understand a codebase or explore an approach | [Ask questions in chat](/docs/chat/chat-overview.md) to get explanations grounded in your code and discuss alternatives. |
| Research a complex change before editing | [Plan with an agent](/docs/agents/run/planning.md) and review the proposed approach before implementation. |
| Make a focused edit without leaving your code | Use [inline chat](/docs/chat/inline-chat.md) to describe the change in the editor. |
| Get code suggestions while you type | Use [inline suggestions](/docs/editing/ai-powered-suggestions.md) for completions and suggested next edits. |
| Complete a common development task | Use [smart actions](/docs/editing/copilot-smart-actions.md) for tasks such as generating a commit message or explaining selected code. |

You use chat both to ask questions and to direct agents. You can ask **Agent** to propose an approach before editing, or select **Plan**, where available, for a dedicated planning workflow.

For a compact comparison, use the [AI features cheat sheet](/docs/agents/reference/ai-features-cheat-sheet.md#choose-a-feature) to choose the least autonomous option that fits your task.

## How you work with an agent

An agent uses a language model to reason about your request and [tools](/docs/agents/concepts/tools.md) to act on your development environment. It gathers [context](/docs/agents/concepts/context.md), such as code, error messages, and test results, rather than relying only on the words in your prompt.

For example, ask an agent to fix a failing test while preserving the intended behavior:

1. The agent finds the test and reads the related code to understand the failure.
1. It edits the implementation or tests based on the task and your constraints.
1. It runs the tests, inspects the results, and revises its changes if needed.
1. You review the diff and validation results, verify the behavior yourself, and decide which changes to keep.

This repeated reasoning, action, and validation is the **agent loop**. You can provide more context, redirect the work, or stop the agent along the way. Learn more about [how agents work](/docs/agents/concepts/agents.md).

The conversation and work for a task belong to a **session**. Sessions keep related context and changes together so you can return to a task or manage independent tasks separately. Learn about [sessions and handoff](/docs/agents/concepts/sessions.md).

## Ways to work with agents

Start with the interface that fits how you want to work. You can continue supported sessions between the Chat view and the Agents window, rather than choosing one interface for every task.

### Work alongside your code

Use the [Chat view](/docs/agents/run/chat-view.md) when you're working in an open project. Keep the conversation beside your editor, inspect changes as they happen, and use the debugger, terminal, and tests as part of the same workflow. This is the starting point for the agents quickstart.

![Screenshot showing the Chat view with the sessions list, conversation, and chat input.](images/agents-overview/chat-view-expanded.png)

For keyboard, screen reader, and low-vision workflows, learn how to use the [Accessible View and other accessibility features](/docs/configure/accessibility/accessibility.md#accessible-view).

### Delegate and manage tasks

Use the [Agents window](/docs/agents/run/agents-window.md) when your focus is assigning tasks and reviewing their results. Manage multiple sessions across projects, follow their progress, and open the editor when you want to work directly with the code.

![Screenshot showing how to start a new agent session by selecting New at the top of the sidebar in the Agents window.](images/agents-overview/agents-window-hero.png)

### Other ways to access agents

For terminal-based work, explore [GitHub Copilot CLI](/docs/agents/run/agent-harnesses.md#use-copilot-cli-from-the-terminal). For work away from your current editor, explore [cloud agents that return pull requests](/docs/agents/run/agent-harnesses.md#start-a-cloud-session) or [remote sessions and browser access](/docs/agents/run/remote-agent-sessions.md). You can also [view supported sessions from other applications](/docs/agents/run/sessions/manage-sessions.md#view-sessions-from-other-applications). The [GitHub Copilot app](https://github.com/features/copilot) provides a dedicated desktop experience outside VS Code.

## Choose your models, agents, and tools

Start with the quickstart's recommended setup, then adjust individual choices to fit your task and project:

* **Models**: choose a [language model](/docs/agents/concepts/language-models.md) based on the reasoning capabilities, speed, and cost your task requires.
* **Agent harnesses**: use a supported [agent harness](/docs/agents/run/agent-harnesses.md), such as Copilot, Anthropic Claude, or OpenAI Codex, for its tools and workflows. The harness connects a model to tools and manages the session, so changing harnesses is different from switching models.
* **Model access**: use models from your GitHub Copilot plan, [bring your own API key (BYOK)](/docs/agent-customization/language-models.md#bring-your-own-language-model-key), or connect a supported local model. These options let you use an existing model provider account or keep model processing local.
* **Tools and customization**: share your coding standards and test commands through [project instructions](/docs/agents/guides/customize-copilot-guide.md). Connect external systems through [Model Context Protocol (MCP) servers](/docs/agent-customization/mcp-servers.md), package recurring tasks as [agent skills](/docs/agent-customization/agent-skills.md), or install [plugins](/docs/agent-customization/agent-plugins.md) that bundle tools and workflows. Compare the options in [agent customization concepts](/docs/agents/concepts/customization.md).

Available models, tools, and customizations depend on the selected harness, your account, and your organization's policies.

Where tools run and where the model is hosted are separate choices. An agent can edit files on your machine while sending model requests to a hosted provider.

## Stay in control

AI can produce incorrect code or misunderstand your intent. You remain responsible for deciding which changes reach your codebase.

* **Review before integrating.** Agents can save edits directly to the session folder or worktree. [Review the diff and validate the result](/docs/agents/run/review-code-edits.md) before you commit or merge changes.
* **Control actions and redirect work.** Use [permissions and approvals](/docs/agents/run/approvals.md) to decide which actions require confirmation. You can [steer or stop a request](/docs/chat/chat-overview.md#send-messages-while-a-request-is-running), but stopping doesn't undo completed actions or changes to external services.
* **Understand isolation.** A Git worktree keeps code changes separate, but isn't a security boundary. For file system and network restrictions on agent-run terminal commands, review the [platform-specific sandboxing options](/docs/agents/run/agent-sandboxing.md).

Before using agents on an existing project, review the [recommended security baseline](/docs/agents/run/security.md#recommended-security-baseline). On a managed device, your organization might restrict available agents, models, and tools through [enterprise AI policies](/docs/enterprise/manage-ai-settings.md). If agents don't appear in chat, review the [availability troubleshooting steps](/docs/agents/agent-troubleshooting/faq.md#agents-are-not-available-in-chat).

## Next steps

* **New to VS Code and AI:** [Install the editor, open a workspace, and choose a first task](/docs/getstarted/overview.md).
* **Already use VS Code, but new to agents:** [Complete a first agent task](/docs/agents/quickstart.md) with a recommended starting configuration.
* **Already use AI agents:** Follow the [experienced-agent fast track](/docs/agents/guides/overview.md#experienced-agent-fast-track) to work in your own repository.
* **Prefer lighter-weight AI:** [Choose an AI feature](/docs/agents/reference/ai-features-cheat-sheet.md#choose-a-feature) for suggestions, focused edits, questions, or delegated work.
* **Ready to improve a project workflow:** [Adapt agents to your project](/docs/agents/guides/customize-copilot-guide.md), then [compare quality, reliability, and usage](/docs/agents/guides/optimize-usage.md).