Build with AI in VS Code

Use AI in Visual Studio Code to understand unfamiliar code, fix bugs, and build features. Get suggestions while you type, ask questions in a conversation, or give an AI agent a task that spans files and tools. You can work alongside the agent in your editor or delegate a task and review the result.

Choose from multiple AI models and agent providers, bring your own model API key, and extend agents with tools and plugins. VS Code brings these options into the same editor, so you can adapt your AI workflow to your task and your team's requirements. You don't need to configure every option before trying your first task.

The quickstart uses GitHub Copilot. See the Copilot setup guide for account, usage, and data-handling requirements, or compare other agent providers and sign-in options.

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.

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.

How you work with an agent

An agent uses a language model to reason about your request and tools to act on your development environment. It gathers context, 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.
2. It edits the implementation or tests based on the task and your constraints.
3. It runs the tests, inspects the results, and revises its changes if needed.
4. 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.

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.

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 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.

Delegate and manage tasks

Use the Agents window (Preview) 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.

Other ways to access agents

For terminal-based work, explore GitHub Copilot CLI. For work away from your current editor, explore cloud agents that return pull requests or remote sessions and browser access. The GitHub Copilot app 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 based on the reasoning capabilities, speed, and cost your task requires.
• Agent harnesses: use a supported agent harness, 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), 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. Connect external systems through Model Context Protocol (MCP) servers, package recurring tasks as agent skills, or install plugins that bundle tools and workflows. Compare the options in agent customization concepts.

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 before you commit or merge changes.
• Control actions and redirect work. Use permissions and approvals to decide which actions require confirmation. You can steer or stop a request, 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.

Before using agents on an existing project, review the recommended security baseline. Your organization might also restrict available agents, models, and tools through enterprise AI policies.

Next steps

• Complete your first task with an agent: build and validate a small app with a recommended starting configuration.
• Follow the agents tutorial: build a portfolio page and learn the agent, editor, browser, and source control workflows.
• Apply the workflow to your own project: choose a bounded task in an existing codebase and review the result.