- oh we should paint it

- we need a paintbrush

- I hear yak hair makes the best paintbrushes

- here I am, shaving a yak

made more sense than the examples given in the op

not intending to slam you personally, but the fact that you finished your projects and then lost interest in them is very indicative of projects done for the sake of generating something with AI just because you could.

I think the people who succeed will be the ones who are willing to develop a project with high effort, and then use the LLM to help with some of that effort. not just because in a world where everyone has access to the same AI code generators the effort is the value add, but because a good product genuinely does require a lot of human input and supervision being l beyond simply churning out code.

then i looked at the code and asked it to benchmark, hinting that it looked like it was doing a lot in the inner loop. and sure enough, adding a few simple graphics to every page more doubled the time it took to generate the largest size of document (~1s -> ~2.2s for ~400 pages). without any more prompting claude figured out that it had an accidentally-quadratic loop, and fixed that.

i then had to tell it "look, we are using a template to avoid regenerating boilerplate with every page. you can add a placeholder to the template and replace it with graphics using xml patching code you already wrote for another part of the doc generation". the final code was a lot cleaner and ran in ~1.2s, which claude (again unprompted, to its credit) did fine-grained benchmarking to prove was the unavoidable overhead of simply inserting all those large chunks of xml into the document.

i wouldn't even say it was a coincidence that i ran into this right after writing my comment about having to micromanage the LLM, because this sort of thing happens all the time. i can say that i had a much easier time doing this because i looked at the code generated in a single commit and could easily see that it smelt off. i would have not have wanted to do this at the end of 20 commits all building on each other.

i would not want to go down the "take myself out of the loop" path because yes, i do have to micromanage the claude session, often course-correcting every commit and then doing large scale refactoring every so often. but i'm perfectly happy doing that - i see claude as more of a tool than a coder i can hand work off to.

on one hand it does seem a bit messed up - this was not in my "job description", so it was technically unpaid work that was nevertheless a formal part of my expectations. but on the other hand I really liked working in an environment where everyone spent some of their time and effort to improve things for everyone.

also making it an explicit requirement for everyone to do at least attempted to circumvent the more toxic culture of "I am a rockstar engineer and I'm busy doing important things, someone else can do the glue work". not to mention that "someone else" usually ended up being a woman, and that she was almost certainly getting paid less than said rockstar engineer.

the OP's implication is that the company should have formally hired someone to do all the glue work, but it is usually made up of enough diverse pieces that it would be practically impossible to hire a single person to do it - e.g. what sort of job title would cover "write documentation, interview software engineers, and organise a team off-site"? but those were all tasks that needed to be done and the citizenship requirement let the burden be distributed more fairly.

I think a better way to put it is not "don't do glue work" but "don't be the only chump doing unrewarded glue work", i.e. to push for a company culture where everyone is expected to do some of it and where it is formally recognised as work.

keeping things simple is not a bad thing, and often requires you to be smart enough not to overcomplicate them.