Recently I’ve noticed general consensus changing towards ai assisted development or as I call it - development. People who got used to it are honing their skills in it. Claude or codex or cursor or any other harness means little, set to frontier model and suddenly everything just works. There are memory lapses and other joys like that but mostly it just works.
Also what changed- I got to ask profs, automated testing, e2e what was 2 years ago trouble to setup (I checked, 2024) now is a bliss. No longer I worry what llm will do, what if I’m not on absolute best one - even cheap and cheerful llms now are plentiful (current like Luna 6.0 or sonnet 5.5). I understand that this post mostly about nothing and it’s ok. My current worries are kind of similar- requirement gathering, choosing right backend, right data shape, contract, setting up ci/cd, e2e, unit testing. Even between these - some tasks are fading into obscurity. Also me stating that it’s right - is a guesswork, usually it’s a discovery. What I like nowadays is getting everything validated and benchmarked where possible - so out of whatever choices I need to make I know what works for me.
I have a few loops like that setup which validate step by step and chooses best for me.
Top comments (0)