A governance model for agentic engineering.
Coding agents are good enough that the bottleneck has moved: no longer how much code an organization can produce, but how much it can be accountable for. This document proposes sixteen concrete operating principles for a software organization in which humans no longer review code — who owns which decisions, what counts as evidence, how changes are verified, and how such an organization stays governable. Part II translates the principles into repositories, permissions, processes, gates and procedures, including what to measure to keep an otherwise opaque harness understandable.
→ Read the paper (~8,000 words, ≈35 min)
The principles are not design aspirations: they run in production. The numbers quoted in Part II come from that installation.
Wolfgang Lubowski — co-founder of TechVera, the EU LLM platform these principles operate. The document grew out of that work; the platform serves as its case study.
Version 1.0.0 — see the changelog. This is a living reference document: argue with it. Issues and pull requests are welcome — disagreements with a concrete failure scenario attached are the most useful kind.
CC BY 4.0 — share and adapt with attribution.