AQ for engineering managers
Published July 28, 2026 · by the AQ team
AQ gives an engineering manager the thing agent tooling usually takes away: visibility. Every agent your team runs lives in a shared workspace you can open like a document: the live terminal, the working tree in an editor, the running app in a preview, and the PRs it produced. No status meeting required; the status is a link.
Your engineers adopted Claude Code and Codex individually, and each one got faster. What you lost in the process is the ambient awareness a team room used to give you: who is working on what, what is stuck, what is about to land. AQ puts that back without adding ceremony.
What you can see
- The workspace list is the live map of work. Every unit of agent work is a workspace with a name, a branch, and a state. Opening the dashboard answers "what is running right now" the way a build wall answers "is CI green."
- Sessions are open-door. A workspace that is visible to the team can be opened by anyone on it. Watch the agent work live, scroll back through what it did, or get the keyboard with a one-click grant from the session owner. No screen-share, no "can you send me the transcript."
- PRs are tracked where the work happened. When an agent opens a pull request, the workspace records it, and merge or close status flows back. You see which agent work actually landed, not just which sessions ran.
Review the behavior, not just the diff
Agent-written code fails in ways human code review is not calibrated for: plausible-looking changes that do not do what the PR description says. AQ keeps the evidence next to the diff. A reviewer opens the workspace, reads the change in the editor, clicks through the live preview of the running app, and pins comments that can be sent back to the agent as its next prompt. Stakeholders outside the team open the preview from a link without an account. The full pattern is in reviewing AI agent work.
Costs you can put in a budget
AQ charges per seat and never meters model usage: agents sign into Claude Code, Codex, and other CLIs with each engineer's existing accounts, and AQ adds no markup on top. The seat price is the whole AQ price, which makes the line item predictable in a way credit-based tools are not. There are two plans: Free, where you bring a VM from your own cloud account, and Team at $100 per user per month (promotional, standard $200) with a dedicated always-on VM included. Details on pricing.
Rolling it out
Start on the Free plan with one repo and the engineers already using agent CLIs; nothing about their workflow changes except where the session runs. Connect Linear if you use it: labeling an issue creates the workspace and keeps the issue status in sync, so the board you already run stays the source of truth. The intake flow is covered in from Linear issue to PR.
Frequently asked questions
How do I see what an AI agent actually did?
Open its workspace. The terminal scrollback shows what the agent ran and why, the editor shows the working tree and diff, the preview shows the running app, and the workspace lists any PRs it opened. The evidence lives in one place instead of being reconstructed from a transcript.
Does AQ replace our code review process?
No. Agents open normal pull requests on your GitHub repos and your existing review process applies. AQ adds the context around the PR: the live session that produced it, the preview to verify behavior, and comments that flow back to the agent.
Can I control who sees which workspace?
Yes. A workspace is either visible to the whole team or private, and a private workspace can be shared with specific people. Sharing is managed by the workspace owner.
How does AQ affect our model spend?
It does not touch it. Each engineer connects the agent CLIs with their own Claude or OpenAI account, so model spend stays exactly where it is today and AQ never marks it up. AQ's own price is per seat only.