GitHub Agent HQ Now Runs Claude and Codex Alongside Copilot

TL;DR

• GitHub Agent HQ now supports Claude and Codex (both in public preview) alongside Copilot for Pro+ and Enterprise users
• Run multiple agents on the same task to compare approaches, surface edge cases, and pressure-test architectural decisions
• All agent work stays inside GitHub and VS Code — no context switching, no copy-paste between tools
• Agent output appears as draft PRs and comments, reviewed like any other contribution

The Big Picture

Context switching kills momentum. You're deep in a refactor, hit a tricky architectural decision, and suddenly you're toggling between your editor, a separate AI chat interface, and GitHub to manually sync everything. By the time you paste the context back, you've lost the thread.

GitHub's Agent HQ solves this by making coding agents native to where you already work. Copilot Pro+ and Enterprise users can now run GitHub Copilot, Claude (Anthropic), and OpenAI Codex directly inside GitHub, GitHub Mobile, and VS Code. No external dashboards. No stateless prompts. The agents operate inside your repository, with full access to issues, PRs, and commit history.

The real shift here isn't just multi-agent support. It's that you can assign multiple agents to the same task and compare how they reason about tradeoffs. One agent might prioritize backward compatibility. Another might flag async edge cases. A third might propose a more modular architecture. You get competing perspectives without leaving your workflow.

This is GitHub positioning itself as the orchestration layer for AI coding tools, not just a host for one proprietary model. If you've been frustrated by vendor lock-in or wanted to test Claude's reasoning against Copilot's speed, this is the first real answer.

How It Works

Agent HQ runs on three surfaces: github.com, GitHub Mobile, and VS Code (version 1.109 or later). Each coding agent session consumes one premium request from your Copilot Pro+ or Enterprise plan. Claude and Codex must be explicitly enabled in your repository settings before use.

On GitHub, you start sessions from the new Agents tab in any enabled repository. You type a request, select an agent using the Copilot icon, and submit. Agents run asynchronously by default. You can watch progress in real time or review completed sessions later. Each session produces detailed logs showing what the agent did and why.

You can also assign agents directly to issues and pull requests. Assign an issue to Copilot, Claude, and Codex simultaneously to compare results. Agents submit draft pull requests for review. You can assign them to existing PRs to request changes or further analysis. Mention @Copilot, @Claude, or @Codex in PR comments to prompt follow-up work.

In VS Code, you open the Agent sessions view from the title bar or command palette. You choose a session type: Local (Copilot), Claude, or Codex for fast, interactive assistance. Cloud for autonomous tasks that run on GitHub, then select an agent. Background for asynchronous local work (Copilot only). This lets you explore ideas in the editor and delegate longer-running work to GitHub without losing context.

Agent activity is logged and reviewable. Their output fits into the same workflows you use to evaluate developer contributions. GitHub is explicit about this: agents can still make mistakes. Their output is designed to be reviewed, compared, and challenged, not blindly accepted.

The architecture here is straightforward. GitHub provides the execution environment, context management, and review infrastructure. Anthropic and OpenAI provide the models. You control which agents run, where they run, and what they can access. Everything stays inside your repository. No data leaves GitHub's infrastructure unless you explicitly push it elsewhere.

What This Changes For Developers

The immediate workflow shift is that you can pressure-test decisions before code hardens. Assign three agents to evaluate a proposed refactor. One checks for modularity and coupling issues. Another hunts for edge cases and async pitfalls. A third proposes the smallest backward-compatible change to keep the blast radius low.

You're not just getting faster code generation. You're getting competing architectural perspectives without scheduling three separate code reviews. This moves your review process from syntax to strategy. Instead of catching bugs after merge, you surface tradeoffs during design.

For teams, this changes how you handle first-pass review. GitHub has integrated a code review step directly into Copilot's workflow. Copilot addresses initial problems before a developer sees the code. This isn't about replacing human review. It's about filtering out the obvious issues so humans focus on the hard decisions.

The enterprise controls matter more than they sound. Admins can define which agents and models are permitted org-wide. GitHub Code Quality (in public preview) extends Copilot's security checks to evaluate maintainability and reliability impact. The Copilot metrics dashboard (also in public preview) tracks usage and impact across your entire organization. Audit logging and enterprise-grade access management ensure agents work with your security posture, not against it.

This is GitHub acknowledging that AI adoption at scale requires governance, not just access. If you're an enterprise team that's been blocked from using Claude because it doesn't integrate with your existing toolchain, that blocker just disappeared.

Try It Yourself

Agent HQ is available now for Copilot Pro+ and Enterprise users. Claude and Codex are in public preview. You'll need to enable them in your repository settings before use.

To enable third-party agents, navigate to your repository settings, find the Copilot section, and toggle on Claude and Codex. Full instructions are in GitHub's documentation.

Once enabled, open the Agents tab in your repository. Try assigning the same issue to Copilot, Claude, and Codex. Compare how each agent structures the solution. Look for differences in error handling, modularity, and edge case coverage. Review the logs to see how each agent reasoned about the problem.

In VS Code, open the Agent sessions view and start a local session with Claude. Ask it to refactor a function for testability. Then start a cloud session with Codex and ask it to evaluate the same function for performance. Compare the results side by side.

If you're already using GitHub's agentic workflows, Agent HQ extends that model to multi-agent orchestration. You're not just automating tasks. You're automating the comparison of competing approaches.

The Bottom Line

Use Agent HQ if you're on Copilot Pro+ or Enterprise and you've hit the limits of single-agent workflows. The ability to compare Claude's reasoning against Copilot's speed or Codex's implementation style is worth the friction of enabling third-party agents. This is especially valuable for architectural decisions where you want multiple perspectives before committing to a direction.

Skip it if you're on a free or individual Copilot plan — you don't have access yet. Also skip it if your team isn't ready to review agent-generated code with the same rigor you'd apply to human contributions. GitHub is clear that agents make mistakes. If your review process can't handle that, you'll just be merging bad code faster.

The real opportunity here is for enterprise teams that have been blocked from adopting Claude or other models because they don't integrate with existing workflows. Agent HQ removes that blocker. The risk is that teams treat agent output as authoritative instead of advisory. GitHub has built the review infrastructure. Whether teams actually use it is a different question.

Source: GitHub Blog