Been testing and evaluating most of software developments principles for long-term maintainability and codebase quality with coding agents.
Feeling so irrating everytime I see a literal 5 comment lines for a function with 1 loop and 2 ifs. Made my own approach to hand over to agents, and so far the benchmarks yielded good results. Sharing with anyone who would want to try or giving some practices on how you burn less tokens with your coding agents, so I can add it to my approach: lessloc
Tbh, since the day of mainly relying on AI fully for implementation, I feel like the less code it produce, the better state the development be in later stages, lots of cases where AI do so much brute-force approachs making context bloating just to understand 3 files.