If you push a lot of new features but your baseline is constantly failing, then something is wrong.
For example, in our company, most commits on main currently have 3-5 authors (we squash): 1-2 humans, 1-3 agents (cursor cloud agent getting started, ppl pulling it into cursor locally to continue, then review using copilot review, modify using copilot agent) then use a vibe coded github app offloading UI test execution to a beefy baremetal machine to adjust baselines.
Copilot review in particular is just so good, better than any agent i know (incl opus 4.7). It just allows you to skip the first few review rounds by humans and fix simple but hard to spot logical bugs, keep docstring & style up to date across the codebase, before you give it to a human - which means everyone can focus on writing more code.
Setting all of this up, at a massive scale, is just not feasible for any of these projects.