[ballenf]

The thesis has it backwards. We will see fewer published/downloaded apps/packages as people rely on others less. I'm not sure we're quite there yet but I'm increasingly likely to spend a few minutes giving an LLM a chance to make a tool I need instead of sifting through sketchy and dodgy websites for some slightly obscure functionality. I use fewer ad-heavy sites that for converting a one text file format to another.

Personally, I see the paid or adware software market shrinking, not growing, as a testament to the success of LLMs in coding.

[meroes]

Ya maybe this. I’ve found some work at the “tool level”. I’m not a programmer, just did RLHF for a few years and AI has helped me make some tools such as a way to scrape and export to excel 35,000 contacts at a company for marketing purposes. Things like that. Yes I know libraries exist and someone who is already a programmer could do this, but also there’s some interesting logic in how to avoid duplicates and interact with modern websites that was impractical for me. And maybe this job is too small for a real programmer.

There are many small, different, and one-time tasks that don’t fit full blown apps. Which I would characterize an AI building a novel app as building a house out of random bits of lumber. It will work but will have no cohesive process and sounds like a nightmare.

[Kalabasa]

Yep. I was looking for a tiling/scrolling window manager for MacOS. Setting it up, learning the configs, reading github issues, learning the key bindings take time. Then I gave up and it was done in 2 days with Claude Code to my preferences. No intention to polish up and publish this tool.