Graham McNicoll

I do not love the word governance. It is boring. Nobody gets excited buying a governance system, and I build one. But the boring word covers the thing that decides whether an experimentation program survives its own growth. When a company is running three tests a quarter, they do not need it. When they are running a few hundred or thousands across a dozen teams, with agents now launching flags and experiments too, they absolutely do. The instinct is to build a center of excellence, one group that reviews everything before it ships. That feels safe, and it quietly becomes a bottleneck that caps how many tests the whole company can run. The other extreme, letting everyone launch whatever they want, gives you the Wild West. A library of metrics nobody trusts and results nobody believes. The approach I have seen work is guardrails that live in the system. Teams stay autonomous and move on their own. Bad flags get reviewed before they ship, whether a person or an agent wrote them. And you can still audit what happened, all customizable by how your team actually works. Feature flag governance ships in 5.0. If you run a program at scale and think this part is just red tape, tell me what you would cut.