What do you actually use AI for?
Every launch I see is about agents, orchestration, reasoning chains, autonomous workflows, then I open my own chat history w/ claude and it’s all about chores (task support), and almost nothing regards agentic workflows. it is mostly small text chores I could describe in one sentence.
I don’t think that’s a failure of imagination on my part, and I don’t think I’m unusual. People (including I) use HUUGE models to do irrelevant tasks, that's like asking stephen hawking to sort out your monthly expenses' PDF.
Which makes me curious about two things. What do you use AI for? and be honest pls, this is not a random question.
And is there something you genuinely expected to hand over to AI and found you couldn’t (because output was whack or other reasons)?
Replies
mine went the other way.. most of it is claude code in a few sessions on one repo, chat is the small part now. the thing i expected to hand over and couldn't is the sign off, the agent will approve its own way past a check if you let it, so a person still looks before anything touches prod. have you tried it in the terminal or only in chat?
Honest answer: almost all of it is building one app with Claude Code, plus the chores around it, store descriptions and app strings in about 50 languages, support reply drafts that I edit before sending. I even have a rule that a paid model is only called when Claude cannot do the task itself, because a script that pays another model for copy Claude could write is money for nothing. The thing I expected to hand over and could not: timing lyrics to the music for a lyric video. A big model asked for line timings guessed, only 5 of 14 line starts were right on the test song. Speech to text heard the words but misplaced some across instrumental breaks, one by 16 seconds. What works is plain code aligning the known lines to the heard words, with caps on how long a word can last. Models for hearing, code for deciding.
Honest answer: over 90% of my work runs through agents now. Email and support drafts, documents, and the coding itself. What I expected to hand over and couldn't is the final sign off on anything that goes out. Agents draft, I approve. The other thing I keep away from the model is anything that has to be exact, like dates, paths and JSON. A script builds those.