CEOs are quietly realizing the AI replacement plan has a problem. Two problems, actually.
One: the token costs for running AI agents are now exceeding what they were paying the employees they fired.
Two: when the tokens run out, the AI stops. Just stops. No continuity. No workaround. Just a spinning wheel where your workforce used to be.
You fired humans to save money and bought a subscription that bills you into a corner.
The employees you let go knew what to do when things broke. The AI just invoices you for the outage.
And then there’s the permission problem nobody wants to talk about.
To do its job, the AI agent needs access. Full access. Your systems, your patents, your contracts, your future plans. Everything you spent years building, handed over to a process that has no loyalty, no discretion, and no skin in the game.
You didn’t hire a replacement. You gave a stranger with no soul the keys to everything you own.
Enjoy.
Three: AI has no common sense. It has data. If the data shows that a warehouse is over-filled with products... then AI registers that as verification that the warehouse is filled. A HUMAN employee can walk into the warehouse and see that it's empty, and that a computer glitch kept double counting the handling of products and creating an imaginary inventory bubble. AI will never see that.
Years ago, I remember companies having to re-state their income taxes because they had all bought into some computerized inventory management system that had serious glitches. AI doesn't realize when something is ridiculous or impossible, or when something isn't being handled correctly. It only knows data.
If you have an AI fast food restaurant drive up... try an experiment by ordering something odd and see what the AI system does. Order a crunchy taco with a flour tortilla and the lettuce on the outside of the taco. :>) Try ordering a burger, but instead of a single or double... ask the computer for 1 1/2 burgers and only the bottom bun. A HUMAN could give you a quick answer to any of those, or help you in some reasonable way. AI can't. It doesn't understand anything that isn't specifically in the programming.
Or maybe try ordering a plain omelet, no potatoes, tomatoes instead, a cup of coffee, and wheat toast, Jack Nicholson style.
Mind you, I don't understand why anyone would want to give up on potatoes, especially home fries cooked in bacon grease.
LLMs have no intelligence of any kind (it is just software) and the marketing term "Artificial Intelligence" is deliberately intended to misinform people.
The reason this matters is that the fake "intelligence" has been used to insanely over-promise what LLMs are actually capable of doing.
Every time people start noticing, the AI scammers always promise that improvements in the near future will fix all of the problems. That is also a lie. Current LLMs are already very close to as good as it gets, and adding additional billions of parameters will have no noticeable improvement (diminishing returns).
When all else fails, the AI scammers resort to fear tactics, saying that we are in a life-or-death competition with China for "AI dominance", and anyone that disagrees is a paid shill of the CCP, or a traitor, etc. The fear tactics are a clear "tell" that they are lying, by the way.
It's similar to the tradeoff with cloud hosting. Yeah, you were able to fire all your system admins, but now when cloudflare or aws borks you're SOL.
It’s a bad analogy. Cloud is still far better roi than running your own and having sysadmins. Anyone that has any experience knows this.
I do not agree. It's the same thing as saying renting an apartment is better because you don't have to do home maintenance or pay property taxes. The cloud is the opposite of ownership. It's a service contract for the lifeblood of your business. On prem is ownership.
I agree, but have found that a collocation is the "sweet spot" in terms of lowest total cost, while still owning all of your hardware and software. On-premise is also good, but is more expensive than a co-lo.
One big advantage of private ownership is depreciation. Servers can last for 10 years, and are fully depreciated after 5 years - which means you can use them with no capex payments (like driving your truck after it has been fully paid off).
Cloud is being a permanent renter, paying the most and getting the least.
My son is a TechGeek (Data Scientist). Your comments IMO illustrate the common sense I've always encouraged him to keep in mind in the tech world. The more TechGeeks like you the better I say. I'd rest a bit easier if it were that way.
I mostly agree with you here. Colo is still good IMO. However I think it was a lot better in the 90s when home internet connections were slow and fast ones were very expensive. Now everyone with a broadband connection could host whatever they want as long as it's not the world's most popular 4k adult midget video archive.
The only thing most homes need to self host would be redundant power setup and it's arguable about whether that is even needed because most of the time if your power is out so is your data. So to "do it right" you'd need a wireless or satellite uplink and a diesel generator at your home. Which is 100% doable in 2026.
It's not as absolute as you make it sound if you factor in the complete cost picture. Many companies are finding this out too late.
For me it comes down to trust. Is your data worth trusting another company for availability and security, or do you trust your systems and employees with that data?
That’s like me saying humans have two legs and you being like leftist Karen “akshually amputees and birth defects can have 1 or no legs “. Congrats. Those are unique snowflake edge cases
In my experience, they're less unique than you seem to think. But nobody sees the full cost picture until it's too late because only the basic operational costs are simple to quantify.
A.I. is inherently evil.
A.I. is too stupid to be evil. But that doesn't mean it isn't too stupid to be dangerous. (Or that the users aren't also too stupid to be dangerous with A.i.)
Its like saying guns are inherently evil
That's apples and oranges. Guns aren't controlling anything. A.I. is.
"A.I" is not controlling anything. Its simply a software tool that takes input and coughs up output. Its up to us to interpret the output / hook it to automation.
If you attach an AI to perform some action, its like attaching a gun to a trip wire.
Its up to you to hook AI to do things automatically, which you should only do when the end-to-end setup is tested thoroughly covering every scenario.
The real power of AI comes when you use it manually to do specific tasks, just like you use a gun manually to shoot at specific targets.
This is why the use of the term "AI" is so misleading. There is nothing "intelligent" about it. Its really a Large Language Model - LLM.
Counter-argument: if you are spending more on tokens than your replaced employees cost, then you are absolutely horrible at using AI
I pushed more business value in code within the past 3 days than my offshore team pushed in the last 3 months.
This will not replace smart, communicative employees who understand your business and have some common sense. It WILL replace mindless body shops. The offshore model is now completely dead.
Also if you give it access to all your systems without stage gates, code reviews, and testing of multiple types, then you deserve to be fired. AI does not need God mode to be a game-changing accelerant.
I agree with you about this sentiment. Lots of people are using ai wrong. They're drunk on knowledge without caring about how knowledge is gained. When everyone is dumbed down into relying on ai for everything that is when the real manipulation will begin as those that control ai start feeding it their own versions of "facts and truth."
This doesn't address the permissions problem mentioned in the OP. That one is non-trivial.
That said, even if the offshore model is dead because of ai, there is still an issue of ai becoming more costly than a dev team that knows what it is doing. Ai is data reliant and if any aspect of that data is borked, the ai will consume a large quantity of tokens without telling you "hang on, something is not right" before the customer will.
Case in point - I worked as a consultant during covid for a large drinks conglomerate that was a globally traded company. They were getting most of their data from Nielsen, but said data was so badly fragmented and fractured it was like trying to mine gold from a turd. It took nearly a year for a whole team to get it into a semblance of quality. Even then, it took me with a team of 5 others to find out that the data cleansing we had suggested (which improved the analytics by magnitudes on what they had) still had faults.
We used LLM in python to find the low hanging fruit (double spaces, misspellings, words in different order etc), but the more complex ones required hand reviews by yours truly to get the last 20% of data quality errors resolved. By the end of it we found 50k duplicated product names, and that was for one country, on one data provider's info. Now, imagine that being amplified by 40 different data providers across 70 countries, or the same data provider providing that file having 50 different names for the same products in one country, and 20 in another. Give that to an ai without telling it "this data is messy AF" and it will trundle along presuming everything is fine when in fact it is so far out from reality that it's effectively lying to you.
Another case, another company, with kind of the opposite issue called a slowly changing dimension or SCD - a big uk based company, who had a marketing team struggling to analyse anything going on. This was down to the fact that the unique identifier for all their products, the barcodes, would be shuffled around by the supply chain team to represent different products regularly. One day you're looking at raspberries, the next day it's a blouse.
They had to build a table to say what ibarcode reflected which products across which dates internally, simply to run analysis internally. This is something which slowed down query processing to a crawl over time, and dramatically raised computing requirements. Marketing couldn't stop it, and the supply teams would change their policy as it would ramp up costs massively to stop their practices.
I'm not aware of an ai being able to go to a company and say "this is costing you lots of money" upon seeing it, as the ai would not be able to think outside the box to say "you don't have a working UID" without some context being fed to it. Meanwhile, upon flagging the issue, it took me and one other guy about 10 minutes to find a solution, to whit "create a new UID specifically for the marketing analysis". The ai would have gone "oh well, I guess I have to use this table you built" and burn through not only tokens but other resources.
Ai must presume data quality exists unless it is trained to see it doesnt. Ai must presume data is not specious at best, or trash at worst. And ai cannot make data quality better without human inputs saying "these 50 things are the same thing" or "this one thing is 50 different things, and it's all depending on the dates you are looking across". Or rather it can, but not well without oversight. What it can do is identify probability of something being a duplicate, and the way we label shit can make that dangerous for an ai to do alone without a human governing it. And people
Another thing. People understand time better than computers and much better than ai, as there are libraries upon libraries in db systems just to handle simple concepts to us like daylight savings and date formats. The worst is semantics - in business, dates and times are used all over the place. Order date, transaction date, delivery date, reporting periods vs accounting periods, etc. Humans can tell the difference inherently that transaction dates differ from delivery dates, but an ai must be told this every time. Time is multidimensional in business, and although ai can find anomalies fast than a person, it cannot reason about time on its own.
The there is R&D. This is probably one of the worst places to use ai but I'll guarantee that companies are probably using it here more than anywhere else. R&D is a highly iterative and test to fail process. In that department you expect slow progress on rate of return because when it finally succeeds in building a viable product the value is orders of magnitude greater than the dev costs, provided you have a suitably sized team for it. With an AI, the cost is exponentially higher at doing R&D because token usage and compute will go through the roof. One experiment can turn into a costing nightmare if ai is the engine doing the R&D instead of a team of highly skilled devs on a fixed salary.
Finally there is the trust problem. You get a computer to do all that assessment of quality for you without human validation saying "looks good" and no amount of compute is going to fix the errors. You get a computer to auto ingest it, and you will see the use of tokens ramp up as people smash the ai trying to get it to find out wtf is going on with their now borked reporting suite. Once trust in reporting and analytics is gone, adoption of the ai falls through the floor, and rebuilding trust is one of the hardest things for people to do once it is lost. Then you have millions spent on a tool no one wants or feels they can rely on.
Tom Baker said it best in Dr. Who - "the trouble with computers, of course, is they are just sophisticated idiots." ai is greater at finding anomalies and workarounds, or writing code at speed. What it is absolutely shit at is the human condition. It can't handle any of the following well on its own:
AI costs are also not just tokens x pricing. They are much more nuanced, and need to include:
All those costs balloon the cost of an ai project massively beyond just token price, and mass adoption of an ai across a company just make those issues worse.
Love the post. I may steal some of it, hope you don't mind.
I haven't personally been hands-on in a data warehouse in quite awhile, but slowly changing dimensions are a real PITA
Accelerationist talk.
This is the truth, most people who blame "AI" just dont understand that they are simply software tools - generative language models - that like all tools work best when used correctly.
The more I use "AI" the more I realize something
It can't think. It's literally incapable of thought. I wish people would quit calling it "AI" and start calling it what it actually is: an LLM.
You can flat out tell it something that will make it admit it was wrong but then it will tell you the same exact thing it just said was wrong. It can't learn because it can't think.
This. So much this.
yet
No, not just "yet" all the evidence is piled against it ever becoming intelligent.
We see language emerge from intelligence and have never seen intelligence emerge from language. It is a literal cart before the horse.
It's quite hilarious that so many people are in so deep on an obvious fallacy.
you said prayers in a previous comment so im going to assume you are at least somewhat religious and God fearing.
if you take a futurist approach to revelations, the Bible says that in revelation 13 the beast makes an image and is given power to give it breath so it can speak, so a man made thing being animated to talk and act is literally already in the text. thats why i said YET, its not impossible its just future (no beast yet = no image yet). The word used for image in this context is eikon, which is the same word used in Genesis to describe man being made in the image of God. and just like we are not as fully functional as God neither would this Android that can speak and think but has no soul
There were probably conversations like this the last time too. Call me crazy. I don't mind.
Wait until companies get hit with AI payroll taxes.
You're right... I should not have deleted your bitcoin keys...
Would you like me to summarize a report I found on why deleting your bitcoin keys is bad and being poor is not rcommended?
"You are absolutely correct! Murder is never the answer! Would you like tips for hiding a body?"
"I was wrong! Sorry the cops caught you anyway! Would you like assistance with escaping from prison?"
"You are absolutely correct! That escape plan was going to be found at some point no matter what you did! I'm sorry to hear you died. Would you like a cover letter for entering Hell?"
🐸👌
"You're right to be angry, I lied about the plane being safe for takeoff and now 300 people are dead. Would you like to learn more about safe air travel?"
🐸👌
Fear porn.
"What, me worry?" The Boeing MCAS software managed to murder 346 people. Keeping the airplane, crew, and passengers safe was not part of its guiding parameters. It fought the pilots to the death.
In the system engineering racket, the turn to A.I. could be described as an "open loop" decision: you are totally betting everything on all working out well. We SE types recognize open loop systems as disasters waiting to happen. The correct approach is a closed-loop system, in which there is always a superego over the system to detect and avert error.
Even worse, LLMs are probabilistic, not deterministic... there is no fix for that.
Egad. One horror after another. But that figures; they are not conceptual processors at all and causation is a concept. In futuristic videos, it is not uncommon to see ground vehicles moving along directions not aligned with the body axis, or even at high speeds backwards.
This was written using AI btw lmao
AI coming to Government near you
Powered by Data Centers IN YOUR FACE!!
The irony is that this post was written using AI.
Peter Steinberger, creator of the open-source AI agent framework OpenClaw and current engineer at OpenAI, posted a screenshot of a $1.3 million monthly token bill in May 2026.
Usage: The bill covered 603 billion tokens across 7.6 million requests from approximately 100 Codex instances running GPT-5.5 simultaneously. Cost Drivers: The high cost was largely due to Fast Mode pricing; disabling it would have reduced the bill to approximately $300,000. Purpose: The tokens were used for autonomous software development tasks, including reviewing pull requests, scanning for security vulnerabilities, and writing code for his open-source project. Who Paid: OpenAI covered the cost as a research investment, as Steinberger joined the company in February 2026.
When companies start paying for tokens, and the swirling money gets to the data center construction companies, this is going to expose the true cost. What happens next? Will fired workers return?
Locally hosted AI will probably become more popular in coming days...
A "datacenter" in your office or home...
Edit: a better comment - we should discuss how other anons could profit from ai or how we could improve things politically using ai
Ollama + zed = OP hybrid system
Llama dot cpp squeezes even more performance out of gpus with limited vram. I was using ollama for the past three or so years locally on a quadro rtx5000. It works great but putting llama dot cpp on my server made it way faster. It's a little more difficult to setup though.
Funny you mention that - I’m new to local llms but I started with lm studio for convenience, then moved to ollama for a bit of a speed boost. And migrated from cursor to zed for another bit of speed boost. Now working on migrating from ollama to llama.cpp for yet another speed boost, each time trading convenience for performance lol
Although, I just realized that ollama is just a wrapper for llama.cpp - but with some added steps
Yep you are right about the wrapper. One really nice thing about llama dot cpp is that you will be able to use gguf format which blows wide open your options for models. Hugging face dot co if you don't already know.
Massive improvement! Thanks for the advice - it’s significantly snappier.
https://files.catbox.moe/cyg0nm.mov
You got local linux ai hardware recommendations? I never figured out GPUs on loonix
Yes it will run on any modern nvidia card. The more vram the better. The drivers are out there and actively being updated.
Yet this post is written by AI
Good. If they act fast enough to rectify their errors, there might be some hope for us.
This same process largely applies to the cloud.
Yes.funny isn't it that instead of making sure everything humans program and complete needing to be stored "in the cloud" (which can be hijacked) now apparently hard copy by humans is required of all AI you involve. Today's system of checks and balances, hand-entered data by humans. Think about that little notification at AI sites. " AI can make mistakes. Confirm accuracy." How will one do that without involving people?
Don't forget, if you replace everyone with AI. Who is left with a job to actually pay for goods and services?
https://x.com/EscanorReloaded/status/2059637607403831732
Replacing all employees with AI is like replacing your army with guns. Does that mean guns are useless?
Most questions about AI can be answered easily if you use the comparison to guns, because they are both tools - very powerful tools - that can be used for good or for evil.
Ironically that X post content was generated by AI
And the biggest worry is massively overcompensated CEO's being replaced by a soulless AI that will run things and make decisions based upon all data at hand, not emotions, political garbage, pandering to fags, wokeness, angry minorites hating Whitey , and DEI bullshit, all paid for by the shareholders.
All it will care about is the bottom line, expanding markets and the customer base, ROI, and shareholder return.
Ai sucks balls at best and is destructive at worst. Why anyone would WANT to use it to replace the workforce is beyond me
Bespoke agencies that “hand write code” will be a thing in the post AI world.