How AQ Runs Its Own Website as a Software Factory

Published November 13, 2026 · by the AQ team

The aq.dev website is kept current by AQ's own bots. A bot spots a question people are asking or a fact on the site that has gone stale, writes a brief, and hands it to an agent that makes the change in its own workspace. The bot then reviews the change and, once it is live, checks the page itself. In the month to mid-November 2026 this loop shipped more than 100 page updates.

AQ is a platform for running a team software factory with AI coding agents like Claude Code and Codex, and AQ runs its own website as one: a bot finds questions and stale facts, an agent makes each change in its own workspace, a second review reads every change, and a person decides anything beyond content.

AQ is where teams put AI agents to work on reviewed pull requests and jobs that repeat, using the agents and subscriptions they already have, on private cloud machines of their own. AQ runs its own website as a software factory: a bot finds the work, an agent makes each change in its own workspace, and every change is reviewed and checked live before the work is closed.

This is a worked example of a software factory. To build one yourself, start with how to build a software factory with Claude Code and Codex.

The loop

Every change follows the same path: something worth fixing is noticed, it waits its turn, it becomes a short written brief, an agent makes the change, automated checks run, a second reviewer reads it, it is released, and the live page is checked. Then the loop picks up the next item.

Who does what

One AQ bot runs the loop. A bot in AQ is a named agent with instructions and rules, and its routines are jobs it does on its own. The bot decides what to work on, writes each brief and starts a work item. Each work item has its own workspace and its own copy of the code, so changes never collide, and one agent makes the change and submits it. The same checks every change to AQ goes through then run, the bot reviews the finished change in Changes, and the normal release process puts it live. Bigger jobs can use a crew, where a lead plans and other agents build and review.

The rules that keep it safe

• Nothing goes live by hand. The only way onto the site is the normal release process, with its checks.
• A person decides anything beyond content. Changes to infrastructure, data, billing, sign-in or security go to a person.
• No secrets in briefs. AQ's secrets reach agents as environment variables and are kept out of transcripts.
• Facts are rechecked every time. Each brief is checked that day against AQ's public docs and pricing.

Keeping up when other companies change

Pages about other products go stale the moment a company changes its plans or limits, so those corrections go first:

• Conductor announced on November 4, 2026 that it was joining Meta. The affected pages were corrected the same day.
• xAI reset Grok Bot's weekly limits for all users on November 9, 2026. The correction to both Grok Bot pages was approved within the hour and live a few minutes later.

Where AQ fits

AQ is a platform for running a team software factory with AI coding agents like Claude Code and Codex, and AQ runs its own website as one: a bot finds questions and stale facts, an agent makes each change in its own workspace, a second review reads every change, and a person decides anything beyond content.

AQ is where teams put AI agents to work on reviewed pull requests and jobs that repeat, using the agents and subscriptions they already have, on private cloud machines of their own. Everything this loop uses is part of the product: bots and routines, work items that each get their own workspace, crews for bigger jobs, review in Changes, and secrets kept out of prompts. The judgment stays with the team: the evidence behind each brief, a reviewer that checks claims, and a person who owns decisions beyond content. To set up your own, start with bots and routines or the build guide linked above.