AQ for agencies
Published July 28, 2026 · by the AQ team
AQ lets an agency run AI coding agents across many client projects without the two things clients punish: work leaking across engagements, and progress they cannot see. Each client's repos get their own workspaces, workspaces can be private to exactly the people on that engagement, and every workspace has a live preview a client can open from a link and comment on directly.
Agencies are structurally the best fit for agent leverage: many parallel codebases, deadline-driven work, and billable hours that agents multiply. The blocker is operational, and that is the layer AQ owns. The workflow deep-dive is in coding agents for agencies and client work.
One engagement, one boundary
Connect each client's repos separately, and give each piece of work its own workspace: an isolated git worktree with its own branch, terminal, editor, and preview. Workspaces for a sensitive engagement can be marked private and shared with only the teammates staffed on it, so the intern on project A never stumbles into project B. Sharing is owner-managed and explicit.
The demo is always ready
Every workspace can run its dev server as a live preview with a shareable link. That converts the weekly status call into something better: the client clicks the link and uses the feature as it stands today, no account needed. Send them a review link that allows commenting and, after a quick email sign-in, they pin comments on the exact pixel they mean. Those comments can be sent to the agent as its next prompt, which means client feedback literally becomes the work order. No staging deploy, no screen-recording.
Custody you can put in the MSA
On the Free plan, AQ runs on a VM in your own cloud account: code, worktrees, and CLI logins stay in infrastructure you control, which is a sentence you can put in a security questionnaire. Agents sign in with your engineers' own Claude and OpenAI accounts, and AQ never marks up model usage, so client billing stays clean: your seats are the only AQ line item. The Team plan ($100 per user per month promotional, standard $200) swaps your VM for a dedicated always-on one AQ operates.
Throughput per engineer
The agency math is throughput: one engineer supervising several agents across two client projects, with all of it visible in one dashboard. Parallel workspaces keep the agents isolated from each other, and sessions run server-side, so the overnight agent run on client A finishes while your team sleeps. See running multiple agent sessions in parallel.
Frequently asked questions
Can clients see progress without an AQ account?
Yes. Each workspace's live preview has a shareable link that works for people outside the team: clients open the running app and try the feature with no account. To collect their feedback in place, send a review link that allows commenting; they sign in with their email in one step, so every comment has a name on it, and they never need workspace access.
How do we keep client engagements separated?
Repos are connected per project, every workspace is an isolated git worktree on its own branch, and workspaces can be private and shared with only the people staffed on that engagement. Visibility is explicit and owner-managed, not inherited.
Whose infrastructure does client code run on?
Your choice. On the Free plan it is a VM in your own cloud account, so custody stays with you end to end. On the Team plan AQ provisions a dedicated always-on VM for your team in its own isolated network; nothing runs on shared infrastructure on either plan.
How does AQ show up in what we bill clients?
Simply. AQ is per seat, and model usage runs on your engineers' existing Claude and OpenAI accounts with no markup, so there is no per-token line that fluctuates with agent activity.