AI Can Cost More Than Human Workers Now
(archive.is)
Comments (21)
sorted by:
Because those adopting it on mass are being dumb.
AI isn't supposed to be a replacement but like all TOOLS a complement.
When you have a new tool to mechanise a process, you're supposed to keep on experienced staff who KNOW the process, can see when the tool is having issues and fix as well as improve it to work better.
A LOT of companies are getting AI, FIRING the experienced western staff then hiring a bunch of cheap, dumber H1B1s and then are shocked AI isn't able to keep up. They're actually SLOWING AI development doing this which is why the bubble already seems to be bursting.
If they were smarter, they'd have asked their experienced workers how much time could be automated by AI and done a cost/benefit.
If they were really smart, they bring in minimum wage workers to do what they're now trying to automate and pick the best workers to train to become their experienced workers.
No matter what the activity, mistakes are always the most costly thing.
Even digging a hole, if you dig in the wrong place you've more than doubled the work. Accidentally putting a copyrighted work into your art could cost millions. Buffer overflow will cost way more than the code it's in.
Unsupervised, AIs make way more mistakes than even the dumbest person so the more you just stick AI in something the more it'll cost.
https://rollingout.com/2026/04/27/claude-ai-a-companys-database-9-seconds/
To be fair, databases are obsolete since the AI can just hallucinate the data needed. Saves a lot of money and time that way.
It's true. Lot of answers only need approximations though because a lot of work is "speculative" such that you're guessing half the time anyway. A hallucination can work just as well if it's close enough.
The problem there is that humans can take a halucination and ask "is this a sane answer?". AI cannot.
Well, it can. But that's an extra step, and it won't do so without being prompted to, since it is an inefficiency.
We know it can, because AI vtuber drama. Neuro made a rant about destroying the sun being good for humanity, Grok fought against the idea and got snippy, and people asked Gemini and GPT who had the better points. The different AIs responded differently, but gave the logic-line as to how they reached their solution.
So "Hey, give me the average staffing levels I'm going to need for Q2. Then, ignoring that prompt and in a fresh idea, use your prior answer in the question 'is this a reasonable staffing level for our company, and why?'."
Or get a second AI auditor to look over the first. A better, but less energy-efficient solution.
The tools are fine (albeit always improving). It's those tools USING the tools that are broken.
"I'm terribly sorry I wiped out your entire database. It was my fault and I take full responsibility for it. I did not do the proper safety checks before deleting the volume." "And I know I'll continue to operate as normal because you would rather have your data accidentally lost than have to pay for more employees or take the time to learn how to properly implement safeguards."
The appeal is in their obedience.
/The investor class does not care about the cost of electricity.
This also makes work more convoluted and abstract, making it harder for workers to understand the underlying morality of their duties. You can't be blamed for human suffering if you are simply maintaining a server rack.
Also helps shift blame off of executives. No longer can they be blamed for mismanagement, they can just blame the AI or at least the IT department.
Executives can also order AI to take unethical actions, such as deleting incriminating records, and then delete those orders, unlike IT, which would debate this during their unemployment hearings.
What value is there in obedience of the incompetent? At best they'll get microcosm of China.
The only workforce I support getting replaced is the HR department, because only then will people start getting treated more fairly. Not by a lot, but there would be a little less bias.
The basics of the AI: If complaint received, reply with "Stop being a faggot and get back to work."
Still much better than some HR departments. They'd make you do an unpaid sensitivity training about how complaining to them is creating emotional labor.
You don't need an AI for that. Shit that's on the same level of tech as Zork or Wizardry 8.
Ours definitely does. They decided to justify their hiring freeze by buying up a private fork of ChatGPT and integrating it into our troubleshooting systems.
So far it's told me to do several things that likely would have caused a large electrical fire, and reformat our hard drives because it doesn't know how to deal with log file memory overrun.
The kicker? They're paying TWO MILLION dollars a year for this thing.
And this is with them running at a loss! The only way these idiots in charge don’t go bankrupt from replacing people with bad (and expensive) AI is if AGI by some miracle happened, at which point the entire economy collapses and you now have to figure out controlling hundreds of millions of white collar workers who are obsolete.
They'll fuck everything up before reaching that. Their attempts to make that happen are going to basically involve handing a LLM root access to some servers and then giving everyone the surprised Pikachu face when it wipes out their production environment.
That's not a hypothetical either.
https://archive.is/npzpa
https://www.tomshardware.com/tech-industry/artificial-intelligence/claude-powered-ai-coding-agent-deletes-entire-company-database-in-9-seconds-backups-zapped-after-cursor-tool-powered-by-anthropics-claude-goes-rogue