Back when I was at Icon we would just sling AI slop into the codebase all day and it got us to 10M ARR. It works for a smaller company and I’d absolutely recommend it for a pre-PMF startup (why would you even look at the code when it would change in a couple weeks?) but it’s been interesting to be on the other side at a larger company where there is a higher bar for correctness, where there must be a human element when you’re building for millions of users and dealing w so many people’s financial data AI still writes most of the code, it’s so good at churning out code, but every line gets reviewed by engineers (+ usually other ais too). a lot more time gets spent on understanding what is going on in these like big AI generated PRs, sanity checking safety/correctness, and making sure things mostly scales nicely with the overall complex system (from a maintainability standpoint as well). compared to just a few years ago, the development bottleneck in healthy mid/larger tech companies has essentially shifted to reading, reviewing and contextual understanding