259
“I’m calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history. Agents cannot program…” (geohot.github.io)
submitted 1 month ago by to c/webdev@programming.dev
83 comments fedilink hide all child comments
 

I’m calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history. Agents cannot program, and it’s taking longer and longer to realize that they can’t. They are a highly sophisticated statistical model designed to mimic the distribution of programming. The output is broken, but in a way that’s getting harder and harder to detect. Which is exactly what you’d expect from an increasingly accurate statistical model.

you are viewing a single comment's thread
view the rest of the comments
[–] -1 points 1 month ago* (2 children)

I have yet to be impressed. I’m not very convinced*. I asked for format type mappings between Pipewire, WebGpu, and Vulcan and both ChatGPT and Gemini failed very badly only providing the most common type mappings. This should be a wildly easy task, something any programmer or even beginner programmers can complete. It’s just very boring, mundane, buffer shifting like work. It almost feels like pencil pushing.

Why couldn’t they do it?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 4 points 1 month ago (1 child)

    Are you using real agents or the free chats on the web, because the latter ones are really dumb. Even when you ask them to search the web for basis you don't get much success.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] -1 points 1 month ago* (1 child)

    I haven’t paid

    edit: Well I did for grok, but not on purpose. Grok still failed too.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 2 points 1 month ago*

    The bigger paid models or potentially the local ones if given access to search the web can probably get you the right answers. The big ones have pretty much memorized half the internet, but can still be wrong so pushing them to verify their answers.

    But the harder part is trusting what they say regardless. I can't just take an answer for truth, and unless I can verify the statement (fact checking myself, looking up the source, running/testing the code, etc) then it gets harder to do anything with AI. This is the thing I hate about AI in general, people just take whatever they say at face value. Lawyers with fake citations, random people asking chatgpt about random facts and such. Its a tool that people put too much faith in to do thinking for them.

  • permalink
  • fedilink
  • source
  • parent