AQ vs GitHub Copilot cloud agent: a job you assign, or a place you work

Published July 19, 2026 · by the AQ team

Choose AQ if agent work should be something your team does together: live shared workspaces (terminal, editor, running preview) on infrastructure you control, running the agent CLIs your engineers already subscribe to. Choose GitHub Copilot cloud agent (renamed from Copilot coding agent in April 2026) if you want to assign tasks from issues and pull requests inside GitHub and get PRs back, billed through your Copilot plan. The one-line difference: Copilot's cloud agent is a job you assign and watch; an AQ workspace is a place your team enters.

GitHub also announced Agent HQ, an umbrella for uniting multiple vendors' agents on GitHub: what GitHub calls "mission control" (a single command center to assign, steer, and track agents) plus a "control plane" governance layer for enterprises. As of July 2026 the multi-vendor part is partial: of five announced vendors, Anthropic Claude and OpenAI Codex are documented as available, in public preview. All claims below are as of July 2026, from GitHub's own docs and changelogs.

AQ vs Copilot cloud agent at a glance

What is Copilot cloud agent?

Copilot cloud agent is GitHub's delegated coding agent: assign it an issue, mention it on a pull request, or start it from the Agents tab, and it researches the repository, plans, and makes changes on a branch in an ephemeral GitHub Actions environment, ending in a pull request (since April 2026 it can also research and plan without opening one). Sessions and their full reasoning logs are visible to anyone with repository access, every commit links back to its session log, and enterprises get a generally available governance surface with audit logging and policy controls. Under the Agent HQ umbrella, Anthropic Claude and OpenAI Codex can run as agents on GitHub in public preview, powered by the Copilot subscription.

What is AQ?

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud.

Every AQ workspace is an isolated git worktree on a VM you control, running the unmodified agent CLIs in tmux, streamed live to every browser that opens it. Teammates watch the same terminal, type into the same session, review the same running preview, and hand work to each other with a URL. Agents authenticate with each user's own Claude, OpenAI, and GitHub sign-ins; work arrives from prompts, branches, or Linear; and every workspace's PRs are tracked for the team.

Where Copilot cloud agent wins

• Native to GitHub. Assignment from issues and PR mentions, sessions in the repository's Agents tab, and commit-to-session-log traceability.
• Team-visible by default. Every session in a repo appears in a shared view with full logs for anyone with access.
• Enterprise governance. A generally available controls surface: org-wide policies, audit logs, and per-vendor agent governance.
• Cheap entry. Included from the $10 per month individual Copilot plan.

Where AQ wins

• Teammates can actually steer. GitHub's docs are explicit: viewers of a session "cannot steer or modify the session." In AQ, entering a workspace means entering the session: any teammate can prompt, interrupt, or take over the same live terminal.
• A place, not a 59-minute job. Copilot sessions are ephemeral Actions runs with a hard 59-minute cap, one branch, and one PR. AQ workspaces persist: the environment, scrollback, running preview, and history are all there tomorrow, from any device.
• Your own model subscriptions. On GitHub, even the Claude and Codex agents bill through Copilot AI credits; you cannot bring the Claude Max or ChatGPT plan your engineers already pay for. AQ runs the actual CLIs on those subscriptions, with no markup.
• The full CLI, not a harnessed agent. AQ runs unmodified Claude Code, Codex, and the rest, with each user's own login and settings. New CLI features arrive when the vendor ships them.
• Previews as the review surface. A designer or PM opens the running app from a link (guest links included) and pins comments that can be sent to the agent. On Copilot, review means reading diffs.

The deeper difference: assignment vs occupancy

GitHub built the best-governed delegation pipeline in the category, and if your team's model of agent work is "file an issue, get a PR," it fits like the rest of GitHub does. AQ starts where that model stops: work that needs a human in the loop mid-run, previews before PRs, sessions that outlive an Actions job, agents compared side by side on the same repository, and a workspace a teammate can be handed rather than a log they can read. Fairly noted: self-hosted runners mean Copilot can execute on your infrastructure too; what it cannot do, by design, is let a second person into the session. That is the line between assignment and occupancy, and AQ lives on the other side of it.

Bottom line

If issue-in, PR-out delegation inside GitHub with enterprise governance is what your team needs, Copilot cloud agent is built for exactly that and priced to be everywhere. If your team wants to work with agents rather than wait for them (live, steerable, persistent, on your own subscriptions and infrastructure) that is AQ, and the two coexist cleanly: plenty of teams keep Copilot for the mechanical backlog and run everything that matters in AQ.