LLM assisted coding exhausts me. Unlike 90% of the people here, I enjoy using...
LLM assisted coding exhausts me. Unlike 90% of the people here, I enjoy using LLMs and think they're on net a good thing. I still type check, lint, and add unit tests, etc. This is a lot of work, it doesn't matter if it is artisanal handwritten or not. Once you start, you need to finish, so I have to keep all this in my head until completion. Rewriting and refactoring still happens and I have to hand diff everything. Bots are only somewhat better at editing documents.
Self-replies
You can't trust them to edit a document, but they can give me a new document that I can diff faster that I can fix it artisanally. The tempo is really fast, if you haven't done it it is hard to explain, sort of like, imagine if your coworkers all got back to you instantly and the ball is back in your court after 20 seconds for everything.
Anyhow, having read the discourse around here, I might as well remind people many things can be true at once
- the llms are useful, they work for me
- some people are not using them well and should get good or stop it.
- non-coding talking heads in the news are in a universe of their own (both the pro and anti LLM sorts)
This is brought to you by about 20 hours a week of LLM assisted programming for the last 6-10 months, using mostly two $20 accounts and a little bit of API calls.
I think the $200 (to $1000s) spend on Claude code surely is mostly wasted tokens 'cause no one is reading the code at all. You need to keep an eye on these bots.