[jerf]

LLMs can clean up after themselves if you steer them. It's not very hard. I often have a two-step dance I do where the LLM first outputs some code and then I prompt it to fix the types up to my standards. I haven't found a way to prompt it with any number of skills or CLAUDE.mds or anything else to get it to do it the way I want on the first pass, but it's not that hard to just fix it afterwards. It's a fast enough process that it feels fine using it. One can even make a case for it being a decent way to operate anyhow; make a sketch, then firm it up isn't entirely unlike my manual coding process was anyhow.

The main problem is that pointing an LLM at a codebase and telling it to just "make it better" taps out pretty quickly. That is, not that there's zero juice to squeeze there, but there's not a ton. You can get a bit of improvement but it also rapidly starts changing things just to change things, which I'm not even going to complain about all that much because there's a sense in which it is simply doing as you asked.

So you still need human taste and direction. This will be especially true for something the size of that codebase where you can only hold small fractions of the actual code in the context window at once. Summaries only get you so far.

[andai]

I asked an agent for some high level cleanup and refactoring. I found that I ended up disagreeing with most of the structural changes. Some of them were necessary, some were beneficial, but largely it turned straight line code into abstract factory manager type stuff.

More broadly I've found that making code more elegant (e.g. by removing duplication) increases the cognitive load, because now you can't just read the code anymore but need to mentally "decompress" the higher level structures and indirection into the straight line code, the "code that actually runs."

[jerf]

If it wasn't clear, I don't ask the agent to "clean up my code". I ask it to "take this map[string]string you pass around with constant keys and turn that into a structure" or "extract this API provider out into an interface and make everything using it use this interface instead" or other concrete instructions.

LLMs, to a first approximation, already did as well as they could on the first pass. You can get a bit more out of them by asking them to just try harder, but not much. Whereas if you go in with specific changes they are pretty good at implementing them.

This especially matters because my personal style deviates from the common practices. This may also be why I can't get it to just happen by prompting for it. But if you walk it through it's perfectly capable of transforming the code into a better style, and it's still way faster than trying to write it from scratch.

[andai]

> my personal style deviates from the common practices.

The closest I could get to mine was by telling it to write like a Unix hacker. Hahaha

[maccard]

> LLMs can clean up after themselves if you steer them. It's not very hard. I often have a two-step dance I do where the LLM first outputs some code and then I prompt it to fix the types up to my standards.

My experience has been the opposite. I prompt it and it generates something that mostly works, but steering it into something that would actually be maintainable is an exercise in futility. GPT-5.5 uses up my entire allocation of tokens in just reading the context docs and doing a single shot task.

> aven't found a way to prompt it with any number of skills or CLAUDE.mds or anything else to get it to do it the way I want on the first pass, but it's not that hard to just fix it afterwards.

I've yet to find a way to get an LLM to actually follow the instructions in CLAUDE/Agents.md It very quickly gets to a point where it forgets explicit instructions such as "run clang-format on all .h/.cpp files that you touch", and saying "our coding standards can be found at <link to Notion/Confluence>, please ensure all code adheres strictly to this standard" is ignored. I'd also say that the first pass of the code is very often not even close to how it _should_ be implemented so it's not just review and patch up, it's rearchitect + restructure 50% of the code.

> it also rapidly starts changing things just to change things,

Agreed. LLM's are (very good) text generators. They are good for generating code, and left to their own devices they will generate and generate and generate. Getting them to edit, simplify, and foresee future problems is something that people keep saying "use the latest model, it's amazing (despite saying that about the last 3 models" or "you just need to use <harness|framework>" or "your agents.md needs to contain XYZ" will solve.

[jerf]

"My experience has been the opposite. I prompt it and it generates something that mostly works, but steering it into something that would actually be maintainable is an exercise in futility."

I've seen others have this experience, too.

It would be interesting to sit us down next to each other for a day or two and compare how we do things, but I suspect much less than that won't reveal much of interest.

It is weird to me how often I have to remind the models about their skills or CLAUDE.md (or equivalent). Though I've sort of taken it as just another way I can impact the process, because sometimes I'm happy for them to forget a particular skill for a moment... particularly when I want it to just do a thing based on my prompt and it decides to invoke the "OK let's design this super carefully" skill. I've gotten a lot of use out of that one but I'm pretty comfortable having to explicitly invoke it.

[maccard]

> It would be interesting to sit us down next to each other for a day or two and compare how we do things, but I suspect much less than that won't reveal much of interest.

I'm inclined to agree. If I had to write a hypothesis, I would say that is' likely that people like me underestimate the capability of the tools, and people who disagree likely underestimate the amount of manual steering and/or overlook the things that I find offensive!

[jasonlotito]

> I've yet to find a way to get an LLM to actually follow the instructions in CLAUDE/Agents.md

I'll give you a hint: do you rely on people to do the right things, or do you have automated unit, integration, and e2e testing? Do you have linters? Do you have static analysis that automatically runs and will block when violated?

If you are verifying humans, why aren't you verifying LLMs?

[maccard]

Yes, and agents are generating code that fails those checks regularly. I wouldn’t accept a junior engineer saying “oh I just submit my code and see if the tests pass, let me know if they’re failing and I’ll fix them”. Pre commit hooks are disable-able and LLMs have shown they’re more than willing to say “tests were failing before this change” when they weren’t, or to use python scripts to self escape their sandbox restrictions.