Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

There are widely divergent views here. It'd be hard to have a good discussion unless people mention what tasks they're attempting and failing at. And we'll also have to ask if those tasks (or categories) are representative of mainstream developer effort.

Without mentioning what the LLMs are failing or succeeding at, it's all noise.



We'd need:

- language/framework

- problem space/domain

- SRE experience level

- LLM (model/version)

- agentic harness (claude code, codex, copilot, etc.)

- observed failure modes or win states

- experience wrangling these systems ("I touched ChatGPT once" vs "I spend 12h/day in Claude Code")

And there's more, is the engineer working on a single codebase for 10 years or do they jump around various projects all the time. Is it more greenfield, or legacy maintenance. Is it some frontier never-before-seen research project or CRUD? And so on.


“Youre holding it wrong”




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: