22
AI Code Is Insane Trash
(www.youtube.com)
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 09 Aug 2026
22 points (89.3% liked)
Hacker News
5289 readers
632 users here now
Posts from the RSS Feed of HackerNews.
The feed sometimes contains ads and posts that have been removed by the mod team at HN.
founded 2 years ago
MODERATORS
I've been regularly looking into the current state of what the big models offer and just last week tried vibe coding some small apps. The code is OK at best, which is rarely the case. If you're using a popular language, popular frameworks and are doing common tasks, the output will probably work. But the code ain't great. It truly is like the people we used to joke about back in the day who copied entire apps from StackOverflow. Snippet by snippet. Because in honesty, that is what they are all doing. They are just going by statistics and it is indeed surprising how far you can get with that. But it completely lacks understanding. There are hallucinations every now and then and the better models catch those by compiling and changing the code until it works. If that doesn't sound great, that's because it isn't. Whenever someone says the code AI writes is amazing and as good if not better as that of a human has no idea what they're talking about and/or is trying to sell AI.
I'm a software engineer with 20 years in the industry. I use Claude every day and the majority of the code I write is generated by it. When you set it up and manage it correctly it writes code of a quality that varies from fine to excellent. You still need to have a human reviewing it but it is just radically more productive than what I could put out myself.
Pure vibe coding will quickly end up with a complete mess. But in the hands of a good engineer that understands how to manage it and can give it explicit technical direction in how to write the code, it usually writes high quality code and is a huge productivity multiplier. Like any complex tool, knowing how to use the tool is a skill in itself. Almost everyone I know in the industry is using it and people who aren't will get left behind (I know that's a cliche but it's true).
There's a lot of problems with AI, ethically, financially and culturally. But people shouldn't delude themselves about its capabilities, that will actually make it more difficult to try and improve some of these problems.
All of this is true. What it omits though is that there are so few of us of sufficient skill to wield them without abdicating thinking to the clankers. Instead, what I see is the mid programmers shit out mid code at an astronomical rate, leaving more foibles and poor quality in the MRs that make it to me. I end up needing to claude-by-proxy more frequently than I like. If you never had ~~curiosity~~ paranoia before clankers existed, you're sure as hell not going to have it now when you can most of the time implicitly trust what they output (eg. Sure, let's do something extra that can throw and log that warning inside a high throughput ingestion pipeline for a very common case, WCGW?). The problem is that "most of the time" is not sufficient for systems that are responsible for vast sums of money and regulatory compliance.
Sure, but you're always going to get people that are bad at their job, AI coding models haven't changed that. The point I was trying to make is that people saying that AI can't be used to write high quality code are just wrong.