Hacker News Reader

Comment by agumonkey | Hacker News Reader

You seem to have experience, I dont mind factoring / unifying logic, when done sensibly with enough history in the trenches. It pains me more whenever a young dev comes in and barks "we must merge these two things!" repeatedly without planning for more than two cases and starting to add more and more boolean variables. Crystal makers. Then the obvious issue comes, the two variants weren't that close and now there's one god class trying to handle all forces in one big state.

I agree that LLMs are naturally anti abstraction machines.. I'm often trying to find way to reverse that.

View on HN

> I agree that LLMs are naturally anti abstraction machines.. I'm often trying to find way to reverse that.

I am a bit of an LLM cynic but I am trying to learn it all, and I have to say I have spent most time trying to work out: how do you explain how a brown-field codebase actually works, in such a way that the LLM won't pervert it through misunderstanding.

It does encourage you towards the "conventional" coding standard for any new project, because you want to use a pattern that it will have seen in its training set.

But for example there are differences of opinion in how wordpress plugins (which have a very complex control flow) should be structured. LLMs are incredible at knowing how WP works, actually, but what is difficult is explaining how your methodology for a large plugin is going to work.

It is a battle — but a useful one because it can be used for, er, studying the comparative belief systems of the LLMs.

They don’t have a useful belief system, one of the rookie mistakes of using LLMs is asking them what you “should” do

Absolutely. I think the bit I still struggle with is finding a way to get them to join my team (which is a team of one very tired person).

A story I like is that in the now lost era of handwriting recognition on PDAs, Jef Raskin concluded that the easiest way to solve the problem was to change handwriting so as to meet the algorithm in the middle.

That is, to find a noticeable simplification of handwriting that people could learn quickly and that eliminated hard-to-process quirks.

I feel I am there with the LLM at the moment, trying to work out what the common ground is.

The gold standard is code samples. I've got 1000-line convention documents with very simple rules like "Early returns on a single line". Llms sometimes ignore these or misinterpret them in unusual ways.

But if I tell it "read these files that use the same conventions" first, there's no misunderstanding, and the agent also picks up the general "tone" of the code. I have very little to tweak if I've defined the problem well.

> But if I tell it "read these files that use the same conventions" first, there's no misunderstanding, and the agent also picks up the general "tone" of the code. I have very little to tweak if I've defined the problem well.

Oh that is a bloomin' great idea, and I can fully see how it might work better.

Can't tell you how valuable this comment has been to me and now I feel so much better about evidently kicking a hornet's nest ;-) Thank you so much.

Glad I could help. I've been trying to use coding agents more than makes sense this year to get a feel for the tech. There's no good set of guidelines yet and everything feels like secret knowledge.

If you're using a coding agent like codex or claude code, I've also seen marked improvement by telling the agent to keep a journal of decision points, and every file read or written. And then, here's the important part, read the last five journals before starting. It primes the context with whatever you were working on and keeps a new session more focused that if it has to go searching for keywords through the whole codebase. It can also be an interesting read.

Reply on news.ycombinator.com