If your product manager is capable of destroying your production data by accident with shitty ai code then you have fundamental infrastructure issues that made such data loss inevitable.
this post was submitted on 25 Jul 2025
243 points (98.0% liked)
Technology
73254 readers
3846 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
They are making the infrastructure with the same shitty AI code…
And/or windows admins.
As a former windows admin, I can attest to the fact that Windows was shitty code well before AI made it trendy to write shitty code.
Nut up and push to main
Everyone knows that an excel plain text database is the best
I'm sure this is the moment that will give pause and encourage caution with LLMs going forward. Yes. This time people will learn.
CEOs: wftlolnope bro we doin live neurosurgery with this bitch tomorrow!
From TFA:
"I have failed you completely and catastrophically," Gemini CLI output stated. "My review of the commands confirms my gross incompetence."
Most accurate thing an LLM has ever vomited up
Oh no we're definitely two years away from AGI.
Well, they train most of these with Reddit data, so judging by the iq of shit posters on Reddit, we are on track.
Also, there is a really slick feedback loop here. Reddit uses ai slop from last year to generate content, next year we train the ai using the posts from Reddit.
Hey, it's good at fake apologies! As good as the CEO, no doubt.
Fucking lmao this is the same level of idiocy as curling random shell scripts and piping them straight into bash
the same? idk i think this is worse… who does vibe coding, sees the huge number of errors AI makes and THEN is like “yeah sure i’ll pipe that straight into a prod DB shell”
random bash scripts from reputable projects at least work most of the time (eg like homebrew)
Again, they were running dev tools in prod.
That's a mistake so dumb, it's actually in the ISO that tells you how not to be a fuck up.
You don't run your dev tools in prod.
You don't, but AI does.
Yeah they seemed to be editing code live in production with no revision control. What morons.
Wait, does this mean the Replit case was actually real!? I was sure it was just a skit or rage bait. I didnt think anybody could be that stupid...
It was real. The story itself certainly gave The Onion a run for their money.
It's a race to the bottom. So far The Onion is doing a great job, but reality is so much better at this.
It had such a GLaDOS vibe, so i thought it was a reference to that. I guess life really does imitate art.
"Eleven times in all caps" is going to be my mental image for vibe coders for a very long time. It sounds too stupid to be real, but, I want to believe.
I didnt think anybody could be that stupid...
Where have you been for the last 5 years?
And can I come wherever that was? Sounds blissful.
it's either real, or the guy is really really sticking to the joke
I would never admit this happened to me
I'm sure the number is higher than 2
Vive coder doesn't even know git push? I think I don't believe it.
I'm convinced the AI models ran across the "best place to test is in production" joke in training and thought it was a valid process.
I mean, it is a valid process. You will get far faster results that way than on a test server.
Watching slopvangelical LLM thumpers slip up will never not be entertaining.
Sucks to be stupid enough to use AI instead of professionals. Especially when they actually trust AI more than any human programmer. Because human programmers code gets checked and verified before it moves to production.
AI has some use but it always needs human oversight and the final decision must also be made by a human professional. If you use AI to speed up tasks and you know whether the output of the AI is valid or not, and you have the final decision, then you can safely use it. But if you let AI decide on and execute important tasks basically autonomously, then you have a recipe for disaster. Fully autonomous and mistake-free AI is a naive pipe dream which I don't see on the horizon at all.
~~A good craftsman never blames their tools~~
EDIT: I misquoted
A bad workman blames their tools
On one hand, it's very clear the tools are the core of the problem.
On the other hand, I don't think I'll be calling any vibe coders good craftsman any time soon.
A bad workman blames their tools
FTFY
Thanks! I edited it
Most tradespeople are responsible for choosing and maintaining their own tools. In a lot of cases it’s the foreman of this metaphor who is making the choice and should be taking the blame.
You say "mistakes" I say "features". That's what they wanted, right? Humans are not accountable, couldn't be helped, that's life.