GitHub Copilot in VS Code v1.109: Multi-Agent Development and Claude Agent Support

VS Code v1.109: GitHub Copilot's January 2026 Release Makes Multi-Agent Development the Default

Released on February 4, 2026, the January 2026 update to GitHub Copilot in Visual Studio Code (version 1.109) is one of the most architecturally significant releases in the extension's history. Microsoft explicitly positioned the update around a single theme: making VS Code "the home for multi-agent development." The release introduces Claude agent support, a new session management layer capable of orchestrating multiple AI assistants simultaneously, and a range of improvements to how agents are customized, secured, and extended.

Claude Agent Support in Public Preview

The headline addition is Claude agent support, now in public preview for GitHub Copilot Pro+, Business, and Enterprise subscribers. The integration uses Anthropic's official Claude Agent SDK — meaning developers interact with Claude using the same prompts, tools, and architectural patterns that native Claude implementations use, rather than a wrapper or approximation.

With Claude available alongside the existing GitHub Copilot agent and the newly supported OpenAI Codex agent, developers can choose which model best fits a given task and run multiple agents in the same session. Claude can be started as a local agent for fast, interactive help or delegated as a cloud agent for longer-running, asynchronous tasks — mirroring the flexibility already available with GitHub Copilot's own coding agent.

Multi-Agent Session Management

The release introduces a new session management system designed to accommodate the increasing complexity of working with multiple AI assistants simultaneously. From a single unified view, developers can:

• Run GitHub Copilot, Claude, and Codex agents in parallel
• Manage local agents (fast, interactive), background agents (asynchronous, in-repository), and cloud agents (long-running) from the same interface
• Hand off tasks between agents or intervene in an agent's work mid-session
• Track the status of parallel subagent execution across sessions

Agent status indicators make it clear which agents are active, idle, or awaiting input — reducing the cognitive overhead of managing concurrent AI workflows.

Agent Customization and Orchestration

Agent Skills, which allow developers to package specialized capabilities or domain expertise into reusable workflows, are now generally available and enabled by default. Organizations can define Agent Skills that encapsulate specific toolchains, coding standards, or domain knowledge and share them across teams.

Beyond individual skills, v1.109 introduces agent orchestrations — the ability to build repeatable, team-specific workflows that combine multiple skills and agents into a structured sequence. This enables teams to codify complex multi-step development workflows (e.g., "analyze the issue, write the fix, run tests, open a PR") into a single triggerable orchestration.

Copilot Memory

Copilot Memory is now available in public preview, allowing the assistant to retain relevant context across interactions and sessions. Rather than starting fresh each time, Copilot can build on what it has learned about a developer's preferences, codebase patterns, and project conventions. Memory is shared across Copilot features, including the coding agent, providing consistency across different interaction modes.

Security: Terminal Command Sandboxing

For developers using agent mode, v1.109 introduces experimental terminal command sandboxing on macOS and Linux. When enabled, the sandbox restricts file system and network access for agent-executed terminal commands, reducing the blast radius if an agent runs a command with unintended side effects. The release also brings enhanced auto-approval rules, allowing developers to pre-approve categories of safe operations so agents can proceed without manual confirmation on routine steps.

Chat and Inline Experience Improvements

The inline chat experience in the editor has been redesigned for more natural in-context interactions. Streaming responsiveness has been improved for faster, snappier output, and reasoning capabilities have been enhanced for more reliable answers on complex tasks. Thinking tokens now provide better visibility into the model's reasoning process — particularly useful when working with models that expose chain-of-thought reasoning.

MCP App Integrations

Richer integrations with MCP (Model Context Protocol) apps enable more tool-driven, interactive Copilot experiences directly within VS Code. Developers can now connect Copilot to external tools and services through MCP, expanding the range of actions the assistant can take without leaving the editor.