some people think they’re good at coding but I haven’t seen it. they make a lot of code but a lot of it is wrong, and you’ve still got (optimistically) the same number of engineers as before to review the output. building the scaffolding to get something useful out of coding agents is time-consuming, the results are brittle, and your chatbot or agent framework vendor can break it at any time.
the only unalloyed wins for llms that I’ve been able to find are:
scripting tasks that require image recognition
language translation
even 30b local models are very, very good at these.
in my experience there’s very little sabotage required, the chatbots aren’t actually good for much. see: https://www.wheresyoured.at/i-will-never-respect-a-website/
some people think they’re good at coding but I haven’t seen it. they make a lot of code but a lot of it is wrong, and you’ve still got (optimistically) the same number of engineers as before to review the output. building the scaffolding to get something useful out of coding agents is time-consuming, the results are brittle, and your chatbot or agent framework vendor can break it at any time.
the only unalloyed wins for llms that I’ve been able to find are:
even 30b local models are very, very good at these.