Autopoiesis: A Concept for Repositories to Ship a Preconfigured AI Platform

Summary

Jakob Breu proposes Autopoiesis, a concept for making a repository carry its own preconfigured AI platform rather than only project-specific skills and context. The idea addresses the difficulty of introducing a consistent AI workflow across engineering teams with different backgrounds and working styles, including corporate environments where not everyone is an AI enthusiast. The platform’s source code would remain outside the repository, while containers, configuration, and scripts would provide the project-specific setup in roughly 20 KB in the unoptimized proof of concept. A user would clone the repository, prepare configuration, run a Docker command, and access the AI frontend in a browser. The frontend can also change the harness settings, creating a limited form of self-mutation. Breu describes the implementation as a very early proof of concept rather than a finished product. He suggests that future versions could ship default skills and instructions and connect to a project’s issue tracker, CI system, test environments, and other services through MCPs.