Hacker News Reader

Comment by unclebucknasty | Hacker News Reader

I still have to remind Ai not to use the n+1 pattern in queries, and fight it to stop using inconsistent inline styles vs classes, even within a single page.

How are people getting this "waste of time not to use AI" and "generates 99% of my code" level of code quality?

I know people's response will likely be something, something, "add instructions to memory to not use n+1", etc. But if the Ai wants to generate code this bad, then it represents an overall quality risk that would require an infinite memory file.

View on HN

I've been developing a suspicion that many people simply aren't looking at what their agents produce and can't speak to the quality of it in a fully informed way.

For context, I've set up harnesses with recursive automated review loops, spec driven development, explicit lists, better models, better harnesses, formal methods tooling, etc. All of it helps, but they don't eliminate output issues. Those become very apparent when I go through the slow, manual work of deeply comprehending / validating LLM code.

And that leads me to one of three conclusions. Either my standards are achievable only by hyperintelligent programming gods, I have a skill issue using LLMs, or others aren't applying the same level of attention.

The first is obviously untrue. I meet my own standards and I'm an idiot. The second seems unlikely because I can see my competent coworkers and well-regarded people in the community discussing the same issues. So that leaves the third.

If you assume that programmers’ skills follows a bell curve, the typical output will not be better than the average. And that leads directly into your third conclusion: average programmers just do not put in the effort. That’s literally part of what makes them average.

Unfortunately most people just treat it as a job. And indeed, it is a job for most people. Not all farmers love gardening and all in their free time. It's just a livelihood. For me I like programming but honestly for my job I can't be assed to do it by hand anymore. It looks good enough, screw it, it's not like they pay me extra to produce handwritten code. When the whole org is pushing down this one path you know it's time to dgaf

It works if you neglect code quality and understanding. I've done a few vibe coded projects on my free time but I would feel shame for submitting that kind of code at work where I use LLMs more responsibly.

> How are people getting this "waste of time not to use AI" and "generates 99% of my code" level of code quality?

They don't know what n+1 problem is and they don't know about long-term maintainability.

tbf many human devs I've worked with don't think about n+1 and similar issues and need constant reminders.

overall software quality seems largely unchanged

>tbf many human devs I've worked with don't...

OTOH, if someone insists on writing bad code—even after being advised otherwise—then they should expect to be out of a job.

Doesn't seem like "terminated junior developer" is the level of quality we should be accepting or promoting for AI.

that doesn't actually happen though. If you fired every dev with sloppy code you wouldn't be left with much.

You previously asserted that devs repeatedly ignored specific code quality instructions, which is clearly a fireable offense. You're now generalizing out to a vague concept of "sloppy code", which is not even quantifiable.

In any case, the point is that low quality code shouldn't be promoted as the acceptable goal for AI.

The truth is people just don't seem care about code quality anymore (if they ever did?). I don't think it generates great code (yes even using the state of the art, frontier, super max pro 9000 turbo boost ++ models), but it usually generates code that works. Sadly, that's all that most of the people in charge care about. I mean shit, half the time it doesn't even have to work great, look at the state of a lot of modern software. People complain about it all the time. Quite sad, but that seems to just be the state of things these days.

I don't think corporate ever gave a shit about code quality. Honestly if there weren't building codes skyscrapers would be made with the cheapest shit that didn't collapse immediately and your house would be made of cardboard. Similarly if users were OK with buggy crap or if they had no choice (e.g. windows) then you can get away with lots of crap.

You could only argue about it when an engineer was in charge and even so when they're high up enough the pressures and incentives are to just ship fast and break stuff. Not every company can afford to be a NASA making uncrashable code that runs for 100 years and goes to pluto and all.

It’s quite an uncanny alignment with Nvidia and other semiconductor companies that benefit from their customers to bloating software and buying as many of their chips with no real thought given to efficiency of the runtime.

Probably the largest part is that this kind of code quality is not that important to a lot of devs and applications. Personally I would not describe the majority of code I work on as particularly hard, the challenge was always in getting what I needed done efficiently as opposed to getting it done at all, and LLMs are extremely helpful for that. Though I think I'm more at 90% LLM-generated code, I still do spend a fair bit of time tidying up layout (and any user-visible prose).

I have the agent (Opus 5) spawn code reviews for their own work, and the code reviews generally catch n+1 queries.

Reply on news.ycombinator.com