How far can the "12 factors" be automated?

[BradKML]

This is kind of a weird question since this repo is more or less about "product ownership", but I feel like for hobby projects, it is better to be slightly loose with the rules. This means that more indy automation can be done "to taste".
There are a lot of tooling for Cursor, ClaudeCode, and RooCode/KiloCode for this purpose, the most recent one being Claude Flow, and yet they are not completely built such that pipelines can be self-optimizing https://github.com/ruvnet/claude-flow/
Why self-optimization? Everyone is working towards emergent design, but they mostly target statistics rather than user preferences. That is the part where scaling human feedback is necessary ("say less" and letting agents do most of the analysis) https://github.com/GAIR-NLP/ASI-Arch https://github.com/codelion/openevolve