Software Dark Factory：让治理适配 AI 辅助开发流程

Too often it means more process, more forms, more meetings, more gates, and more delay. Red tape dressed up as control.

That is not what I’m trying to build with Software Dark Factory.

The point of SDF is not ceremony. It is to help teams keep useful delivery moving while making AI-assisted work easier to review, trace, and challenge.

The danger with AI agents is not just that they can write bad code. It is that they can produce a lot of plausible-looking work very quickly, without leaving enough evidence for a reviewer to understand what changed, why it changed, what was checked, and what boundaries were respected.

So the SDF approach is simple: governance should fit the development workflow, not take it over.

The work should still be anchored to a useful, bounded change. The reviewer should still be looking at a recognizable PR. The evidence should be close at hand, without turning the PR into a compliance document.

That is why we are dogfooding the SDF front door around a practical operating question: how quickly can we get from a small governed change to a reviewable draft PR, with the evidence present and the reviewer experience still recognizably normal?

If governance improves trust but kills momentum, real engineering teams will route around it.

The goal is controlled speed. Not process for process’s sake. Not AI theatre. Not a compliance wrapper stapled on after the work is done.

Just a better front door for AI-assisted delivery: useful changes, normal PRs, visible evidence, and reviewer confidence without needless drag.