[lucidone]

I find AI useful, I use it most days to write snippets of code or to rubber duck with. It hasn't changed my workflows that much, just replaced Stackoverflow with ChatGPT. Feels like the sweet spot for me, everything else is noise.

[barbazoo]

Chat is the obvious application but the real value imho is using LLM to bridge a gap non-deterministically you couldn’t bridge deterministically before. Entity extraction for example allows us to connect two workflows that often required a human in the loop. Not anymore. I see this everywhere in our SaaS product.

[variadix]

For the majority of the questions asked and answered on StackOverflow, LLMs are undeniably better. I’m thinking of things like ‘How do I do X in Python’, ‘How do I do X on Linux’, etc., questions that are small in scope, not open ended, and can be easily verified. For everything else LLMs range from rarely useful to outright misleading and counterproductive.

LLMs also don’t really enable coincidental discovery like search engines do. Having to RTFM or read a spec or a book or a blog post to figure out the answer to your question sometimes also teaches you about related and important concepts that you wouldn’t have come across otherwise, and usually there will be suggestions for further reading or a side bar with other interesting topics etc. Completely replacing search feels like a a bit like a trap, where what you get in immediate answers you lose in an unseen opportunity cost.

[jimbokun]

Yep, replaces Google, Stackoverflow and autocomplete for coding with a much superior experience.

But anyone taking a vibe coded project with no human understanding of the code produced and puts it straight into production is going to have a bad time.

[bwfan123]

same,

- a better summarizing google for some queries

- snippet generator

but has not changed any workflows.