Well, I don't see how linters/test on a final cluster is bad. This is why we have unit/integration/alpha/canary tests. The first 1 can happen on my laptop over 5 seconds while the second can happen on my personal instance over 20 minutes. Usually first means the second will pass and I just need to run it in the background without breaking any workflows.

i.e. It might not be a bad feedback loop, you just need more of them that are faster.