The Living Repo

Imagine reporting a bug, going to lunch, and coming back to a tested fix waiting for review. Creating the issue started the work. An agent reproduced the problem, fixed it, and brought you the evidence.

Now imagine that being the ordinary way your project works.

This is The Living Repo: a repository organized so AI can operate the software development lifecycle. Your goal as a developer is to build the machine that builds the machine.

An app is never really finished. Customers get confused, discover bugs, and ask for things you never considered. You release another version, learn something, and repeat. The app is one moment in a continuing process.

That process deserves engineering attention of its own.

Give the repo the tools to act

My preferred home for it is a monorepo. Put product code beside tests, build scripts, release procedures, and issue records. Use AGENTS.md files to explain local rules, skills to describe recurring work, and standard and custom CLI tools to make that work repeatable.

Connect outside services too: GitHub, Jira, Monday, customer support, hosted builds. Give information a path from customer report to verified change to release.

AI makes these pieces work together. A trigger starts an agent; the agent reads the context, uses the tools, and works toward the goal. You define success and decide where human judgment belongs. Each completed attempt leaves knowledge for the next.

Every time you copy instructions between windows or explain the same procedure again, you have found another piece of the machine to build.

Let it improve its own process

Now give it this issue: “It takes too much work to get a bug fixed.”

The same agent that fixes your app can improve the process that fixes your app. It can automate test setup, repair incomplete instructions, or make review evidence easier to evaluate. Once accepted, those improvements help future agents.

Now the machine is improving the machine that builds the machine.

One bug fix improves one behavior. A better reproduction tool can improve the next hundred fixes. Your engineering work accumulates as a growing ability to do engineering work.

Start with one recurring problem

Document what success means, provide the tools, connect the trigger, and review the results. Improve whatever keeps requiring your intervention.

You still need customer understanding, taste, and priorities. A machine that closes issues quickly can build something nobody wants. Keep customer feedback in the loop.

I think this is the future of software development: repositories that carry work forward and improve their own ability to solve problems.

Build The Living Repo.