AQ vs Claude Code on the web: same agent, different room

Published July 19, 2026 · by the AQ team

Choose AQ if your team should be in the room while Claude Code works: live shared terminals, a code editor, and app previews that any teammate opens from a browser, running on infrastructure you control. Choose Claude Code on the web if you want to delegate well-defined tasks to Claude in Anthropic's cloud from claude.ai or the mobile app, solo. The one-line difference: it is the same agent on the same Claude subscription either way; what changes is whether the session is a private chat-and-diff surface in Anthropic's sandbox or a persistent multiplayer workspace in your own cloud.

This is the rare comparison where the model bill is identical: both run on the Claude plan you already pay for. Everything else about the room differs. All claims about Claude Code on the web are as of July 2026, from Anthropic's docs, and it is labeled a research preview there.

AQ vs Claude Code on the web at a glance

What is Claude Code on the web?

Claude Code on the web (claude.ai/code) is Anthropic's cloud surface for Claude Code: connect GitHub repositories, describe tasks, and Claude works in isolated Anthropic-managed VMs, with automatic PR creation and parallel tasks across repositories. The owner can watch and steer the session, review diffs with inline comments, and pull a session down into the terminal (Anthropic calls it teleport). Anthropic's own guidance positions it for well-defined tasks that do not need frequent steering, and directs bigger or more interactive workloads to the terminal or to running Claude Code on your own hardware. Sharing is a visibility toggle plus a link; per Anthropic's docs, recipients see the latest state when they open it but "their view doesn't update in real time."

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.

In AQ, Claude Code is the same unmodified CLI with the same login, running in tmux on a VM you control, inside a workspace that is a place rather than a conversation: an isolated git worktree, a real terminal your whole team can watch and type into, a code editor, and a live dev-server preview with shareable guest links and comments that can be sent to the agent as prompts. Work arrives from prompts, branches, or Linear, and PRs are tracked per workspace for the team.

Where Claude Code on the web wins

• Zero infrastructure on the plan you have. No VM to think about, and no separate compute charge; usage shares your existing Claude rate limits.
• Session portability into the terminal. A cloud session's conversation and branch can be pulled into the CLI and continued locally.
• Credential-isolation engineering. A proxy design keeps git credentials outside the sandbox and restricts pushes to the working branch, with configurable network access levels.
• PR automation extras. It can watch a pull request and respond to CI failures and review comments, and scheduled routines can run without your machine.

Where AQ wins

• Your team is in the room. AQ sessions are multiplayer: a teammate opens the workspace and is in the actual session, typing into the same terminal. On Claude Code on the web, share links are views that do not update in real time, and only the owning account steers.
• A workspace, not a chat. AQ gives the full surface: live terminal, code editor, running app preview with guest review and comments-to-agent. The web surface is chat, diffs, and a PR button.
• Your infrastructure, your ceilings. AQ runs on your VM or a dedicated AQ-managed one: your hardware sizes, your network, your data boundary. Cloud sessions run in Anthropic's sandbox at fixed resource ceilings and are reclaimed after inactivity.
• Every agent, not one. The same AQ workspace model runs Codex, Cursor Agent, Kimi, Grok, Pi, and OpenCode next to Claude Code, so teams compare and mix agents on the same repository.
• Team-level plumbing. Linear intake with two-way sync, ownership following assignees, per-workspace PR tracking, and guest links for people outside the team.

The deeper difference: delegation vs habitation

Anthropic built a very good delegation surface: hand Claude a well-defined task, check back, review a diff. AQ is built for everything that model leaves out: the tasks that need steering, the runs a teammate should sanity-check while they happen, the preview a designer should see before the PR exists, and the handoff that should be a URL instead of a writeup. The two are not even mutually exclusive: the same Claude subscription powers both, and teams that live in AQ still fire off solo cloud tasks. But when the work is a team's work, the session has to be a place more than one person can occupy, and that is the thing AQ is.

Pricing compared

Claude Code on the web is included with Claude Pro, Max, Team, and Enterprise premium seats as of July 2026, sharing your plan's rate limits with no separate compute charge. AQ is $100 per user per month at promotional pricing (standard $200) with a dedicated managed, always-on VM included, plus a Free plan on a VM you bring. On both, Claude's model usage bills through the subscription you already have.

Bottom line

If you want to hand Claude solo tasks in Anthropic's cloud, Claude Code on the web is already in your subscription and does that well-defined-task job as designed. If your team wants to run Claude Code together (live, steerable, reviewable, on infrastructure you control, next to every other agent CLI) that is not what a chat-and-diff surface is for, and it is exactly what AQ is: the multiplayer workspace for the same agent you already pay for.