That is risk management. As the bottleneck moves downstream, we'll need to review our risk appetite. Right now I don't trust AI enough to ship code without a human reviewing it, but if this becomes a massive constraint, you can allow automatic merges for low-impact areas like non-critical internal tooling, as long as you trust the agentic harness you've built.
If there is a sea of changes coming in the way of the HITL, who do we know what is worth reviewing and what is fine to let flow as is?
That is risk management. As the bottleneck moves downstream, we'll need to review our risk appetite. Right now I don't trust AI enough to ship code without a human reviewing it, but if this becomes a massive constraint, you can allow automatic merges for low-impact areas like non-critical internal tooling, as long as you trust the agentic harness you've built.