Architecture as Truth
SystemDox implements Architecture as Truth
Architecture as Truth (AAT) is a practice for building software with AI agents, set out by Emanuele Pugliese in 2026. It needs no product. SystemDox is one implementation, for teams whose truth is spread across many repositories.
The practice
Architecture as Truth (AAT) is a way of building software in which the architecture documents (the decisions, the contracts, the behaviours and the plan) are the single source of truth about the product. People decide by changing those documents, AI agents build from them, and everything that ships is reconciled back into them.
Its five principles and four rules for an AI agent are on the standing reference, architectureastruth.org, with a starter repository and two pipeline checks a team can use today. The practice was first published in the article AI Promoted Everybody. Architecture as Truth (AAT) Is the New Job.
What SystemDox does for each part
In one repository, AAT needs little more than a file the agent is told to read. Across dozens, the hard part is getting the same documents to every person and every agent at the same moment. That is the part SystemDox does.
| AAT asks | SystemDox |
|---|---|
| Write the truth down, in words the whole team can readPrinciple 1 | SystemDox reads the decisions, contracts, scenarios and plans from your team's own Git repositories, which stay the source of truth. Publishing commits to GitHub, so nothing becomes a second copy. |
| Decide in the documents, before the buildPrinciple 2 | Each change's brief (the decision, the contract, the behaviours and the plan) is rendered on a change page for the person who approves it, so they approve documents, not diffs. |
| One truth for people and for agentsPrinciple 3 | The MCP server gives Claude Code, Codex, Cursor or any MCP client the same files a person approved, at the moment the agent writes. |
| Read before writingRule 1 | get_context returns the decisions, contracts and behaviours that govern the task. get_change tells a build session whether its brief has been passed, so it can stop before it writes. |
| Leave the truth trueRule 4 | report_decision drafts a decision record for a person to review, update_spec writes back to a spec, and SystemDox flags the pages the code has moved past. |
What it does not do
Two of the four rules belong to the agent and the pipeline, and SystemDox does not enforce them. Build what was approved: nothing in SystemDox stops an agent from editing an approved contract or the wording of a scenario. The starter's approved_guard.py refuses that change in CI. Stop, do not patch: no tool can make an agent ask instead of guessing. It is asked for in the agent's instruction file, and the record of the change is what shows whether it was kept.
A team can practise AAT with a Git repository, plain text files, a test runner and the open starter, and no SystemDox at all.
Where it runs today
PuglieseWeb builds its own products this way, on SystemDox: more than sixty repositories, one owner who approves, and AI agents that build. On 5 November 2026 its decision repository held 226 decision records and 2,975 dated change records, each naming the documents it reconciled.
Interest declared. SystemDox is built by PuglieseWeb LTD, the company of AAT's author. It is in private preview: it runs our own engineering every day and is not open to the public yet. Request early access and we will write to you personally when a place opens.
Practise AAT across many repositories
The documents stay in your repositories. SystemDox gets the same ones to every person and every agent, at the moment each of them needs them.