AI Was Supposed to Replace Workers. It’s Not Working

Key Takeaways

So, something is happening in AI. Companies are actually returning to humans and scrapping replacing workers with AI. And more information is telling us that this is likely to be the trend for the rest of 2026. The companies that swung the hardest at full automation are now actually walking back the layoffs, switching the bots off, and they're calling humans back to the desk they told they would never need again. Now, this is actually happening at a variety of companies. Starbucks, Klarna, McDonald's, IBM, Air Canada, and dozens more. Replacing humans with AI simply isn't working, and the truth is now too loud to ignore. So, essentially, this started with a Starbucks because the story is the cleanest. I mean, in May 2026, a few days ago, Starbucks killed its big AI inventory system. The system was called Nomad Go, and this was supposed to use cameras and computer vision to count every item in a store automatically, so that no humans would ever have to do inventory again. 11,000 stores ran it, and after a full pilot, Starbucks pulled the plug. And the reason was very simple. The AI got the counts wrong, and humans had to recount everything by hand anyways. So, the company is going back to manual counts done by baristas, and the promise was zero human effort, but the reality was actually double the human effort because workers were now doing their job and the AI's job. Now, you have to think about this logically, okay? The guy, Brian Niccol, who took over the company as a CEO in 2024, he actually pitched investors on a turnaround called Back to Starbucks, and AI was at the center of that turnaround. The story was that smart cameras and predictive software would handle inventory, scheduling, and forecasting, so the company could run with fewer and fewer hands, and investors loved it. Wall Street analysts wrote it up as the future of retail. The very first big AI deployment under that plan is in the bin. The CEO has not announced a replacement strategy. Company has gone quiet on AI specifically, and is talking about giving baristas more time with the customers. Now, what's crazy about this, okay? And this is where we have this problem in this industry that I've seen a time and time again, is that the Nomad Go demos worked in a controlled environment and with perfectly lit shelves and clean product. The AI hit 99% accuracy, and that's what Starbucks bought. But real stores are messy. Coffee bags get stacked sideways, syrup bottles get half hidden, lighting changes, and once systems meet the real world, the accuracy collapses. And workers reported that the AI constantly miscounted and they had to recount every shelf to fix it. The replacement plan died because the demo and the deployment were two different products. And you see this happening in AI all the time. How many times have they released a different model from the one that they demoed in those videos, and it simply cannot do the same thing? I mean, this is the unfortunate reality when it comes to working with AI. Often times, what you're finding out is that unfortunately, AI actually doesn't save time in many industries, it adds time. And at Starbucks, a barista's job was supposed to shrink because the AI was counting for them, but instead, the barista had to count anyway and also fix the AI's mistakes and explain to the manager why the numbers did not match. And this is the time tax of fake automation. The company buys the software, pays for the software, the worker still does the work, and then the worker does the extra work on top to clean up after the software. That's not replacement when you think about it, that's just replacement theater. Now, when you think about this, okay, people are finally starting to wake up to this, okay? After this Starbucks news broke a couple of days ago, analysts started to write that AI as a labor replacement strategy is now being repriced inside earnings models. For the last 2 years, any company that announced AI-driven layoffs got a huge boost in their stock price. The market just assumed that those layoffs would stick and that the savings would flow straight to the bottom line, but the problem is is that now the market is seeing layoffs reverse, AI tools get switched off, and rehiring is quietly beginning. And that financial story is shifting from AI saves money to AI costs a lot of money, and then we're going to have to hire everyone back anyways. Now, the most important number in this whole story is coming from MIT. So, in August 2025, MIT researchers, you probably heard about this before, they ran a study called the Gen AI Divide, and they found something pretty staggering, and that's that 95% of generative AI pilot programs at large companies were failing to deliver any measurable revenue or cost impact. Not bad results, just no results at all. And almost every replacement project, every chatbot, every automated workflow sat there burning money without moving a single business number. And only 5% of pilots actually worked. The other 95% were quietly burning the company money. Now, this is interesting because of course studies are running concurrently, and we are going to see more studies come out in 2026. But one of the things I think we should pay attention to is one of the most famous ones, which is of course Klarna. And that was one of the the most famous AI stories, okay? The buy now, pay later company. In 2024, the CEO Sebastian went on every podcast, every magazine cover, and he was just saying, "Look, Klarna has essentially built a customer service agent that did the work of 700 humans." And they actually froze hiring across the entire company and said that AI could do every job at Klarna. Their headcount dropped from over 5,000 to around 3,500, and he was the poster child for the replace everyone with the AI thesis. Every other CEO on Earth was being asked when they were going to do their Klarna moment. But the thing is is that in 2025, Klarna quietly reverses. Customer satisfaction crashed, the AI agent couldn't handle easy questions, but anything complicated, anything emotional, anything where a customer actually needed help, it failed, and people hated it. They called for a human and got a bot, and they got the same wrong answer 15 times, and then complaints started to pile up, and Klarna started hiring humans back, calling them gig customer service workers, and then started to route the hard questions to them. And then they actually had to admit on stage that they'd pushed too far on cost-cutting, and that the quality had suffered as a result. The poster child for AI replacement just became the poster child for AI rollback. Now, and so the worst thing about this entire scenario is that like AI replacing the easy part actually creates a bottleneck on the hard part because now the humans who are actually working there, they're only encountering the cases where the humans are super frustrated. So, they're going to get burned out and then the customer service scores are getting worse, not better, even after hiring the humans back because those humans are now only drowning in calls. I mean, this chart clearly explains it clearly where AI can do 60 to 70% of jobs, but it just isn't there for that, you know, 30%. Now, this is not the only story. We had McDonald's removing their AI-powered ordering technology from their drive-thru restaurants in the United States. And whilst this one is, you know, in early 2024, this is still happening across many different chains. So, McDonald's tried this same playbook and in a partnership with IBM, they ran they ran AI voice ordering, and they actually started started earlier in 2021. Now, the pitch was that AI could take your order faster than human, never get tired, and of course be there all the time. But 3 years and millions of dollars later, McDonald's just decided, "You know what? We're killing that product." And, you know, I'm pretty sure you've maybe seen videos of viral customers screaming at the AI as it added nine iced teas to one order as it refuses to understand a Big Mac, and as it tries to charge people for orders they never made. And, of course, drive-thrus are now back to humans, but that replacement experiment ran for 3 years, ended with the original workers walking back into the headsets. And, like I said, this wasn't just McDonald's. This has happened across many different industries, but the problem is that those edge cases still aren't covered by AI. AI isn't there yet. And you have to remember, okay? A lot of times people are going to say, "Well, you know, AI is going to change and stuff like that." But, remember, guys, that this is essentially built into how these AI models are. So, it's going to be really difficult to remove these things in the future. Hallucinations are part of the model. And when you talk about hallucinations, Air Canada tried to replace its customer service with an AI chatbot and they got sued. So now, not only are the hallucinations bad, they're opening people up to litigation. In 2024, the chatbot told a grieving passenger he could get a bereavement discount after the fact and the airline tried to argue in court that the chatbot was its own legal entity and the airline was not responsible. The judge said, "No, Air Canada was forced to pay." The story became the legal precedent that if your AI replacement tells a lie, your company is on the hook for that lie. >> [music] >> That is a big precedent, okay? Think about it. Companies are now going to have to think about the fact that since these LLMs hallucinate even a small percentage of the time, do they want to have to continually pay out in those small cases where they're going to be on the hook for whatever that chatbot hallucinates? That is pretty crazy, okay? Think about it. Companies are going to be thinking twice whether or not they want to be using these chatbots because is that cost saving going to outweigh the decision of potentially getting sued or having to pay up because your AI agreed to some ridiculous claim. Now, one of the most damaging admissions from 2026 has come from Uber. In December 2025, Uber rolled out Anthropic's Claude code to 5,000 engineers. They built an internal leaderboard to gamify usage. Adoption exploded and by April 2026, Uber's Chief Technology Officer told The Information that the company had burned through its entire AI coding budget in 4 months. Individual engineers were running up between $500 and $2,000 a month each on AI tokens. 70% of Uber's code now originates with AI and yet in May 2026, Uber's own Chief Operating Officer went out there and on the podcast, which I'm about to show you guys, admitted that he could not draw a line between the AI usage and any actual improvement in features shipped to customers. And his exact words, which you're about to see, is that it just isn't there yet. >> Uh our last quarter were AI driven um or you know, our token usage went from X to Y, or percentage of employees who, you know, all all these sort of numbers. Um, and it's amazing, and I think it's like this massive transformation of society, but then you sometimes go and you talk to your senior engineering leaders, and you're saying, "Okay, how many projects that were on the cutting room floor got moved above the line because of the, you know, productivity gains? Because 25% of our code commits were via Claude Code last last quarter." That link is not there yet, right? Like you you're not I mean, I think maybe implicitly there there's more that is getting shipped, but it's it's it's very hard to draw a line between one of those stats and, "Okay, now we're actually producing like 24.5% more useful consumer features, right?" And and that line is hard to draw. And I think over the over the coming quarters and years, like maybe that will become clearer, but I think today it's hard even if some of the underlying metrics are like trending in a really astronomical direction. >> And now, you want to know something crazy? In May 2026, the company Microsoft that invested $13 billion into OpenAI and another $5 billion into Anthropic, they actually banned the Claude Code for its own engineers. And the crazy thing about all of this, they were told to cancel all of Claude Code licenses after the engineers were reportedly using it too much. And so, the real reason this actually happened was because it's just really expensive. Now, I'm I'm going to be making another video about this, but essentially most people don't realize as well is that even the AI that is effective, things like Claude Code, it is remarkably expensive. So, when you have almost a trillion dollar I'm not even sure if Microsoft is a trillion dollar company. I'm pretty sure they are at this point, but when you have a company of that size that says, "Wait a minute, Claude Code is too expensive," that is pretty crazy when you think about it. If they can't afford it, who can? The company that sells AI to the world told its own people to stop using AI because they could not afford it. And I found another cost here that just goes to show that many of these agentic systems, even though they might be effective, they're still more than humans. Here, an Nvidia VP essentially says that their, you know, token usage actually cost more than their team. And if that's the case, why would you actually start to use these models if they're more expensive? The whole point is to save money here. >> I spoke with a VP at Nvidia who first flagged this to me. He said, "Oh, yeah, for months our costs for my team have been more for AI than humans." So, that was the first flag. And then we started to hear this coming out in droves. Uber's CTO said he already blew out his whole budget for 2026 just on AI-related costs. And obviously, that means he's spending more on that than he's spending on human workers. And now I'm starting to hear especially from startup founders, they're bragging about their AI bills being high because the kind of >> Now, do you want to know the craziest statistic of all of this? Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027. So, this is the crazy scale of it. This is the crazy scale of this. 40% of agentic AI projects canceled by the end of 2027. And one of those key reasons is escalating costs, as we've spoken about, unclear business value, or inadequate risk controls, according to Gartner. Now, when it comes to inadequate risk controls, I actually just found this tweet. So, you can see it says, "CEOs are quietly realizing their AI replacement plan has a problem. Two problems, actually. One, the token cost for running AI agents is now exceeding what they were paying the employees they fired. And two, when those 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, you fired humans to save money, and bought a subscription that builds you into a corner. The employees you let go knew what to do when things broke, and 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, and 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. And all of this is true. The tokens are super expensive now, and when the tokens run out, you don't really have someone to manage them if you're just trying to replace them. And even if you do, you're going to actually have to pay for someone who understands AI, which is not only expensive, but you also got the token costs on top of that. And then of course, to do the job, the AI agent needs access, like full access. And there are so many cool use cases, but because you can't give an AI full access to your systems, those use cases, I do wonder if they will ever truly be fulfilled. Now, I think companies need to take a look at what IBM has done because they're the only company that has seemingly been doing this well. And they never tried to replace humans with AI in the first place. So, IBM, they built two internal AI tools called Ask HR and Ask IT. Ask HR handles 94% of routine HR inquiries, and Ask IT cut IT service interactions by 70%. And here's the main thing people are missing. They didn't lay anyone off. They redeployed the savings into hiring more engineers and sales people. Head count went up, and the CEO said that AI augments humans, it does not replace them. And IBM is essentially the rarest company right now because a company that used AI to win and didn't pretend that humans were the problem. And I think that's what the companies need to start realizing. Whilst it sounds good to just replace humans with AI and just maximum profits all the way to the moon, that doesn't work in reality. AI simply isn't there yet, and using humans in an augmented fashion is going to be 10 times better than just replacing them. Here you can see it says, "Bradford said that while AI is better at tasks, humans are still better equipped to be strategic partners to the business and help unlock people's true potential." And that is the real lesson from every failure in this video in one sentence. AI can replace a task, but it cannot replace a job. A job is a bundle of dozens of tasks, and only a few of them are automatable. The barista is not just counting inventory, she's reading customers, calming complaints, training new hires, fixing the espresso machine, and noticing when something is wrong. The customer service rep is not just answering questions, she's reading tone, building trust, and handling the case the scripts never anticipated. Companies that confused the tasks with the jobs are now the ones writing the rehire emails, and companies that understood the actual difference are the ones actually winning. So, now, when you think about if you're trying to use AI for yourself, what you need to be able to do is know AI a lot, like you need to verify everything. You can't just trust the flashy demos when they come on stage and say, "Hey, this new AI tool can do XYZ." Don't trust that at all. You need to be able to test it and verify it, okay? You need to run the pilot in your worst examples, not your best ones. You need to measure the time it takes for you to fix those mistakes, and then you need to add the time of the cost of the AI. And if those numbers still work, only then can you deploy it, okay? And if it doesn't, then just walk away. You're simply wasting time. And most of the companies in this story, they did not do that. They just bought into the demo and the hype, and then they got the disaster. And so, now, we're in this essentially cleanup phase of the AI replacement era. Boards are quietly re-asking CFOs how much AI inventory they have on the books that are not earning any earning anything. CEOs are firing the AI consultants who promised them the replacement Utopia, and HR teams are rehiring the workers they were told they would never need again. And Gartner has predicted that more than 40% of this stuff is going to be canceled by 2027. So, the replacement story for now seems to be over. And now, here's an interesting thing that nobody had considered. Now, I was browsing Twitter and I came across this, which is essentially breaking news because it was only a few hours ago, and it talks about the fact that Senator Elizabeth Warren just urged to tax the AI to give free services to people. So, essentially, what she's saying in this video, which I'll show you now, is that the wealth creation from AI is going to be so crazy that we need to actually impose some kind of AI tax so that those displaced by it actually can still benefit. I mean, it's pretty crazy when you think about it. Now, when you're going to be deploying it in a company, is there going to be an AI tax when you're using those goods and services? It's going to be really interesting for the future. >> Only scary. Tech execs are warning that AI could lead to a level of wealth concentration that will break society and create a permanent underclass. Those are their exact words. I refuse to accept that. There is no doubt that we need to regulate AI and consider bigger and bolder options to rein in the technology. But understand this, if we're going to build an AI future that works for everyone, then we need to tax AI and invest in people. Taxing AI raises the money we need to deliver universal health care. So, if millions of workers get fired because of AI, those workers don't go bankrupt just from a visit to the doctor. And if AI transforms the future of work, then taxing AI means we'll have the resources to invest in things like free college and apprenticeships programs and a jobs guarantee so that all Americans can have good paying work. That's what taxing AI promises. And here's what it could look like. Right now, companies pay taxes on their workers and get tax breaks for investing in AI. Woah, it's time to make corporations pay their fair share and make sure they're no longer incentivized to fire workers and replace them with AI.