[dotancohen]

Do you have any idea how many times I've written the same text parsers or API glue for different clients? The software field has this same issue - I would imagine that any "knowledge worker" field would. That's probably why ChatGPT is so good at producing code.

How many identical prescriptions do chemists write? Doctors write diagnosis'? Teachers write end of semester summaries? Architects sketch rooms? I'm sure that 80% of some fields' output is near-identical to previous work.

[nmfisher]

I feel exactly the same way - a few years ago I was lamenting the fact that I was writing JSON serialization/deserialization code for the 1000th time, and begging for a tool that could automate it for me. I haven't tried ChatGPT for that specific task just yet, but I'm pretty sure it's exactly the solution I had in mind.

[klabb3]

I mean this sounds much more like a job for tooling and higher level meta-programming. If “everything but X” is identical busywork, then just automate the “everything” part. Using AI to generate code seems equally bad for maintenance and possibly worse for silent errors.

That said, there is a lot of busywork in programming, especially refactoring, adding parameters and stuff like that. Sometimes it’s ok to have such work be “mostly correct”, eg if you have strong type safety and unit tests to catch minor errors.

[mattclarkdotnet]

The thing is, with code you mostly do want the verbose output, with a prescription you do need the output to follow an official format, etc. Tools like these seem to fill a niche where you know what you want, but the target format/process is onerous.

[Spivak]

I think what makes code frustrating is that we absolutely could consolidate all this duplicate boilerplate but the “writing code is easier than reading it” problem makes it so the cheapest path for any individual is duplicating work ad hoc rather than making something reusable.