There are a LOT of people who simply don’t bother to check the output of Claude (or <LLM_OF_CHOICE>).
Source: I’m dealing with an absolute firehose of vibe coded PRs from our contractors right now. I swear to god I’m just conversing with their Claude processes. They are very clearly not only not checking their output, but also just responding to PR comments with Claude. I’m getting very close to just making PR comments as if I was directly addressing my own local Claude process, including feedback loops and doc cross checks and so on, which they’re clearly not bothering with. They’re not even vibe coding well.
I have personal reasons for sticking around. But the biggest ones at this point are how shit the job market is currently, and the fact that I get a sabbatical next spring. After that hits, assuming the job market recovers a bit, I’m planning on fucking off.
Honestly you should they’re literally uploading your comments into claude and copy pasting claudes answer back to you. If your company is efficient a lot of thwm are gonna get laid off and you can just run ten claude agents at once instead of paying ten morons a salary to have claude do their job for them.
They were also using who-knows-what LLMs in a biotech context when we have specific regulations against using general use/commercially-available LLMs. We have rules against that for IP and HIPAA reasons. I’m genuinely baffled at how they didn’t get into Big Trouble for that.
The landscape has shifted dramatically with the promise of big profits from higher productivity / labour cost reduction. Corpos used to be pretty strict with policies against putting proprietary code anywhere external. Now it’s all like - yeah shove it in all these third parties’ mouths, including sometimes competitors, it’ll be part of your perf review. It’s crazy. I don’t think any of this is going back to the previous posture. The profit growth promise is too great.
Hey, if they want to give me a deeply stupid metric to achieve and tie it to comp - like, oh I don’t know, SLOC, or token usage - I am perfectly happy to game that system. And I have done in the past.
I think I’ll have to start that soon. Apparently teaching and leading a bunch of juniors to deliver better than their pay grade isn’t as valuable. I guess they think they can get rid of the juniors and have me do the same with cuckoding.
I mean… you can’t vibecode well, nobody can, your example here well illustrates that.
Vibecoding -> the magic machine operates via special incantations, and that makes me a magician!
AI-assisted coding -> Ok this hallucinating genius dumbass ram devouring monster is simultaneously very capable in some ways and horrendously stupid in others… if I am very gentle and precise with it, as one would be with a toddler, it may be very helpful, but I can never assume it may not have a critical and fundamental incomprehension of some kind.
LLMs are clearly useful, in the right hands.
They’re great for a senior or mid level dev to use as an extremely fallible assistant.
They’re good for junior devs to both learn some basics about code and also some basics about how LLMs are not magic.
Like, plug a junior dev into an LLM and then the ask the dev to review the resulting code.
That’s not a bad way to train up a junior dev.
… But this shit isn’t production ready just because it involved an LLM at some point!
Its literally all experimental!
C Suite and Managers are just fucking idiots!
It is just that simple!
They don’t want to do their jobs anymore, and they fell for marketing hype.
I’m genuinely less mad at dumbass but eager vibe coders caught up in a delusion… they’re disproportionately inexperienced, they wouldn’t really have a way to know better.
The dumbasses in charge that are cramming the kind of workflow and workforce setup you describe down everyone’s throats?
These fucking idiots should know better.
They basically could not be more perfectly incompetent, and they should be roundly laughed at and openly mocked as the idiot clowns they are, to their faces, all the time, everywhere, including work settings.
Literally the survival of the human species may depend on our ability to point out and identify to these people’s faces that they are in fact complete fraud morons.
But also, at the same time, build so many AI datacenters that they would require a second Germany worth of power production to run.
Also the US has essentially zero capacity to natively build new power plants and transformers, we would have to import almost everything needed, and we’re also doing a tariff tradewar with the entire planet.
And, 90% of companies report 0 or negative productivity returns from using AI.
…So basically, constrict energy supply, increase energy demand, and use this energy to mostly just literally directly create massive heat islands that do basically nothing for the actual real economy.
The plan is literally ‘throw gasoline on the fire, and maybe the gasoline will become sentient and tell us a less stupid way to do things.’
… this is a long way of me saying ‘yes i agree with you’.
There are a LOT of people who simply don’t bother to check the output of Claude (or <LLM_OF_CHOICE>).
Source: I’m dealing with an absolute firehose of vibe coded PRs from our contractors right now. I swear to god I’m just conversing with their Claude processes. They are very clearly not only not checking their output, but also just responding to PR comments with Claude. I’m getting very close to just making PR comments as if I was directly addressing my own local Claude process, including feedback loops and doc cross checks and so on, which they’re clearly not bothering with. They’re not even vibe coding well.
Does their manager not ask them to explain their code? Won’t they get fired if they have no clue what they’re submitting?
My manager is unfortunately a big LLM kool-aid drinker. And yes, it’s infuriating
Do the clients not verify if the output is working before accepting it?
I am the one not accepting their work due to how they’re doing it, and it’s getting extremely annoying
How are they getting retained if their work is getting rejected so often?
Don’t ask me, that’s above my pay grade
If the people above your pay grade aren’t better than you, there is no scope for growth.
Don’t I know it.
I have personal reasons for sticking around. But the biggest ones at this point are how shit the job market is currently, and the fact that I get a sabbatical next spring. After that hits, assuming the job market recovers a bit, I’m planning on fucking off.
So you’re saying you’re becoming a problem for business velocity. 🤔🤔🤔🤭
Honestly you should they’re literally uploading your comments into claude and copy pasting claudes answer back to you. If your company is efficient a lot of thwm are gonna get laid off and you can just run ten claude agents at once instead of paying ten morons a salary to have claude do their job for them.
Might be using something cheaper than Claude. Contracting agencies aren’t known to pay a lot.
They were also using who-knows-what LLMs in a biotech context when we have specific regulations against using general use/commercially-available LLMs. We have rules against that for IP and HIPAA reasons. I’m genuinely baffled at how they didn’t get into Big Trouble for that.
The landscape has shifted dramatically with the promise of big profits from higher productivity / labour cost reduction. Corpos used to be pretty strict with policies against putting proprietary code anywhere external. Now it’s all like - yeah shove it in all these third parties’ mouths, including sometimes competitors, it’ll be part of your perf review. It’s crazy. I don’t think any of this is going back to the previous posture. The profit growth promise is too great.
Hey, if they want to give me a deeply stupid metric to achieve and tie it to comp - like, oh I don’t know, SLOC, or token usage - I am perfectly happy to game that system. And I have done in the past.
I think I’ll have to start that soon. Apparently teaching and leading a bunch of juniors to deliver better than their pay grade isn’t as valuable. I guess they think they can get rid of the juniors and have me do the same with cuckoding.
I mean… you can’t vibecode well, nobody can, your example here well illustrates that.
Vibecoding -> the magic machine operates via special incantations, and that makes me a magician!
AI-assisted coding -> Ok this hallucinating genius dumbass ram devouring monster is simultaneously very capable in some ways and horrendously stupid in others… if I am very gentle and precise with it, as one would be with a toddler, it may be very helpful, but I can never assume it may not have a critical and fundamental incomprehension of some kind.
LLMs are clearly useful, in the right hands.
They’re great for a senior or mid level dev to use as an extremely fallible assistant.
They’re good for junior devs to both learn some basics about code and also some basics about how LLMs are not magic.
Like, plug a junior dev into an LLM and then the ask the dev to review the resulting code.
That’s not a bad way to train up a junior dev.
… But this shit isn’t production ready just because it involved an LLM at some point!
Its literally all experimental!
C Suite and Managers are just fucking idiots!
It is just that simple!
They don’t want to do their jobs anymore, and they fell for marketing hype.
I’m genuinely less mad at dumbass but eager vibe coders caught up in a delusion… they’re disproportionately inexperienced, they wouldn’t really have a way to know better.
The dumbasses in charge that are cramming the kind of workflow and workforce setup you describe down everyone’s throats?
These fucking idiots should know better.
They basically could not be more perfectly incompetent, and they should be roundly laughed at and openly mocked as the idiot clowns they are, to their faces, all the time, everywhere, including work settings.
Literally the survival of the human species may depend on our ability to point out and identify to these people’s faces that they are in fact complete fraud morons.
I think we will face environmental collapse before they stop drinking the kool-aid
Currently, US policy is effectively:
Destroy 1/4 of the world’s access to oil.
But also, at the same time, build so many AI datacenters that they would require a second Germany worth of power production to run.
Also the US has essentially zero capacity to natively build new power plants and transformers, we would have to import almost everything needed, and we’re also doing a tariff tradewar with the entire planet.
And, 90% of companies report 0 or negative productivity returns from using AI.
…So basically, constrict energy supply, increase energy demand, and use this energy to mostly just literally directly create massive heat islands that do basically nothing for the actual real economy.
The plan is literally ‘throw gasoline on the fire, and maybe the gasoline will become sentient and tell us a less stupid way to do things.’
… this is a long way of me saying ‘yes i agree with you’.