Transcription
Ever since the release of Chad GBT, AI executives in the news media have been hyperventilating about how AI will take everybody's jobs and thus cause mass unemployment. We are now in the fourth year of the AI bubble and there is very little evidence that AI is displacing a significant number of jobs.
According to the consulting firm Challenger, AI was responsible for 55,000 layoffs in the US in 2025. While that sounds like a lot, we need to put that number into context. In 2025, there were over 1 million layoffs. So, AI was only responsible for 5% of total job losses. For those 55,000 layoffs, that just means the company said the layoffs were due to AI. There's no way to confirm that the jobs were actually replaced by AI.
In many cases, a company needs to conduct layoffs for other reasons. For example, if they overhired in the past, instead of admitting the real reason, they will often credit the layoffs to AI because this sounds better to stock market analysts. In the few cases where companies have actually tried to replace employees with AI, the results have often been poor. A recent study in the UK found that 55% of companies who replaced workers with AI now regret the decision and in many cases are hiring them back.
Of course, different jobs will have different impacts from AI. One of the jobs believed to be most at risk is software engineering. On paper, software engineering should be one of the easiest jobs for AI to replace. LLMs are trained by scraping huge amounts of data from websites like Stack Overflow. Anthropics Claude Code and OpenAI's Codeex can output huge amounts of code very quickly.
In January of 2026, Anthropic CEO Dario Amade made the following prediction. "I have engineers within Enthropic who say, 'I don't write any code anymore. I just I just let the model write the code. I edit it. I do the things around it. I think I don't know. We might be six to 12 months away from when the model is doing most maybe all of what SWES do end to end. And then it's a question of how fast does that loop close.'"
According to Amadeday, within 6 to 12 months, AI will be able to do everything that software engineers do end to end. That might sound scary to aspiring software engineers. In fact, many universities have reported a decrease in computer science majors as students are scared that software engineering jobs will all be replaced by AI. The only problem is Daario Amade made the exact same prediction almost 12 months ago.
In March of 2025, Amade made the following prediction. "But now getting to kind of the job side of this. Um I I I do have a fair amount of concern about this. Um on one hand, I think comparative advantage is a very powerful tool. If I look at coding programming which is one area where AI is making the most progress um what we are finding is we are not far from a world I think we'll be there in three to six months where AI is writing 90% of the code and then in 12 months we may be in a world where AI is writing essentially all of the code that was in March of 2025 in 3 to 6 months AI will write 90% of all code in 12 months which is basically now AI will write essent essentially all of the code."
Of course, this hasn't happened. Basically, since the release of Chat GPT, tech CEOs have been saying that AI is going to take everybody's jobs within 12 months. 12 months later, AI hasn't taken any jobs. But then they make the exact same prediction delayed by another 12 months.
In this video, we'll explain why AI is not replacing a significant number of jobs and is not likely to do so in the foreseeable future. And more importantly, we'll look at why AI CEOs want you to believe that AI will take your job even when it won't. Let's start off by looking at software engineering in particular.
It is true that many software engineers use AI tools. According to Google, 90% of technology workers now use some form of LLM during their work. This may very well be true, but just because software engineers use AI doesn't mean they can be replaced by AI. So, how do coders actually use AI?
Let's say you're a software engineer. Even if you're very experienced and capable, you will forget certain functions from time to time. Back in the day, you would go on Stack Overflow. You'd find examples of the function you're looking for and copy paste that into your own code. AI models are trained by scraping the internet. They can pretty easily recreate everything you would read on Stack Overflow because they are trained on this data.
Now, instead of going to Stack Overflow, you can go to ChatgBT to look up functions. This is much easier. You can even use an AI native IDE like cursor or codeex to have the AI write whatever function directly in your code so you don't need to copy paste. Using AI can save you a little bit of time compared to using Stack Overflow. Most AI coding tools are pretty cheap compared to the salary of a software engineer. So it makes sense that these tools have been widely adopted.
If you see a statistic that 90% of software engineers use AI, this can give a highly misleading impression of how useful the AI actually is. A software engineer does much more than just memorize a bunch of functions from Stack Overflow. In real world environments, engineers must collaborate closely with product managers, designers, security teams, and other developers to define requirements, and make thoughtful trade-offs. They spend significant time debugging issues, maintaining and refactoring existing systems, reviewing code, writing tests, and ensuring reliability and performance at scale.
You can try to use AI to create an entire piece of software or even an app. This is called vibe coding. The idea is that someone with limited or no coding experience can create a new app just by giving the AI natural language prompts. But there are huge problems with this approach. There are very few examples of software or apps that were created with AI alone. The highest profile example I could find was Moltbook.
Moltbook is a recently released social media platform. It has a very basic interface similar in some ways to 4chan. is supposed to be a social media platform for AI agents. The idea is you create an AI agent with chat GPT or Claude and give it access to Moltbook. People's AIs will talk to each other on Moltbook and have completely fake AI conversations. There is zero practical use case for Moltbook. It's basically just an AI gimmick.
The creator of Moltbook is a guy called Matt Schlick. He claims to have used AI to create the Moltbook website. He didn't write a single line of code by hand. Within days, hackers found vulnerabilities in Mlebook's code, which they were able to exploit. Hackers were able to steal 35,000 email addresses and 1.5 million Open AI and anthropic API keys. These were the API keys people used to power their AI agents. If your API keys are stolen, the hacker can use it to rack up huge bills that will be charged to your account. So yes, you can vibe code an entire app or website, but it will have so many security holes and other problems that no real company will be willing to put their name behind it.
Security is a huge problem with vibe coding. The cyber security firm Apiro has witnessed a massive increase in security breaches over the past year. They attribute this to software engineers increasingly relying on AI tools to write code. AI coding tools cannot fully replace software engineers, but maybe they can increase the productivity of each software engineer, so you need fewer of them in total. While this sounds like a reasonable hypothesis, it is not borne out by the data.
Many software engineers think that AI increases their productivity. The AI spits out a lot of code very fast, so it feels like you're working faster. But in reality, the code has so many problems in it that you spend more time fixing the problems than it would take to just write the code by hand.
In early 2025, a nonprofit organization called Model Evaluation and Threat Research, or METR, ran an experiment to see how effective AI coding tools are. They hired 16 software engineers to complete various coding tasks. The software engineers had about 5 years of experience. Half of them were allowed to use AI coding tools while the other half were not. The coding tool was Cursor Pro using Claw 3.7, which was considered to be the best coding LLM at the time. The results were shocking.
Before completing the task, they asked the software engineers to estimate how long it would take. The ones who were not allowed to use AI thought it would take two hours to complete the task. The ones who were allowed to use AI thought it would only take them an hour and a half. In reality, the software engineers who were not allowed to use AI were able to complete the tasks in about 1 hour and 40 minutes, less than they forecasted. The ones who used AI took well over 2 hours. While software engineers think AI speeds them up, it actually slows them down.
Over the past few years, the media hype around AI job losses has reached a point of near absurdity. Let's watch a clip of Anthropic CEO Dario Amade speaking to CNN in May of 2025.
"Dario, you've said that AI could wipe out half of all entry-level white collar jobs and spike unemployment to 10 to 20%. How soon might that happen?"
"Well, let's uh uh uh well, first of all, thanks for having me on the show, but uh uh just to back up a little bit, you know, I've been building AI for over a decade, and I think maybe the most salient feature of the technology, and what is driving all of this is how fast the technology is getting better. Um, a couple years ago, you could say that AI models were maybe as good as a smart high school student. I would say that now they're as good as a smart college student and and and sort of reaching past that. I really worry particularly at the entry level that the AI models are are you know ve very much at the center of what what an entry level human worker would do. Um and so it's hard to estimate you know exactly what the impact would be and and you know there that there's always this question of adaptation and and you know these these technology changes have happened before but I think what is striking to me about the that this this AI boom is that it's bigger and it's broader and it's moving faster than anything has before and so compared to previous technology changes I'm a little bit more worried about the labor impact simply because it's happening so fast that yes people will adapt but they they they may not adapt fast enough and so there there you know there there may be an adjustment period."
"you are running an incredibly important company uh in AI uh and you know this better better than anybody or as well as as you know all the names we know people Sam Alman and others who are working Elon Musk and AI why are you raising the alarm because it's not necessarily I would think in your best interest because a lot of the messages we hear from at least publicly from you know some AI CEOs and stuff is is a little bit more calming saying like, you know, these agents are going to be great in your life and and yes, there may be problems. Uh but, you know, writ large, this is a fantastic thing."
">> Yeah. I mean, you know, I think I think the reason I'm raising the alarm is that I think others others haven't as much and you know, I think I think someone needs to say it. You know,"
Dario Amade did something very clever here and he completely fooled the CNN anchor. Anthropic is developing AI. Amday is sounding the alarm about the negative impacts AI will have on the job market. Supposedly, it will cause the unemployment rate to increase to 20% within the next 5 years. Emay seems credible because he is sounding the alarm about AI. Despite being the CEO of an AI company, his alarmist statements seem to be motivated by a genuine concern about the people who will lose their jobs. While I can't get inside Amadeay's head, I believe his motivations to be very different.
Anthropic mostly sells his cla chatbot and claude code product to enterprise customers. For a company to spend large amounts of money on clawed code, they need to believe that this technology will allow them to lay off software engineers. That's the only way to justify the spending. For Anthropic to sell its products, it needs people to believe that AI can replace human workers. But as we will see shortly, there is very little evidence that AI can replace human workers. In fact, there's a lot of evidence to the contrary.
One of the problems with reporting about AI is that public company CEOs often exaggerate how much their companies are using AI because they think this sounds good to Wall Street analysts. For example, in June of 2025, Salesforce CEO Mark Beni off claimed that AI is doing 30 to 50% of all internal work within Salesforce. If this were really the case, they should lay off 30 to 50% of their employees and recognize huge cost savings. Of course, this hasn't happened.
In September, it was reported that Salesforce laid off 4,000 customer support workers. Salesforce has over 70,000 employees, so this is only about 3%. It was later revealed that Salesforce didn't even lay off 4,000 people. They reassigned 4,000 customer support workers to other roles, but there were no net layoffs. They tried to get AI to do customer service, but this has caused huge problems with faulty responses. A Salesforce executive told the information that the company has lost confidence in LMS over the past year due to poor performance at customer service tasks. Mark Beni off made huge claims about AI which generated a lot of media headlines, but there's really no evidence that Salesforce is getting any efficiency gains from AI.
Another example is Microsoft. If any company can replace workers with AI, it should be Microsoft. Microsoft owns 27% of Open AAI and uses OpenAI's technology to create its own AI coding tools such as GitHub Copilot. Microsoft requires its employees to use AI. AI usage is tied to performance reviews. If you don't use AI enough, you'll have your bonus cut or you might even be fired. In May of 2025, Microsoft CEO Saja Nadella boasted that 20 to 30% of code at Microsoft is now written by AI. The shift to AI coding has coincided with huge problems with Windows 11, which is plagued by bugs and security issues. Despite bold claims about AI replacing humans, Microsoft's number of full-time employees has continued to increase.
Perhaps the most egregious case of AI hype comes from Black Rockck, which is the world's largest asset management company. Black Rockck is a major investor in AI infrastructure. In 2024, they created a new subsidiary called the Black Rockck AI Partnership, which invests tens of billions of dollars into AI data centers. Black Rockck CEO Larry Frink is a huge proponent of AI and claims that his company is deploying AI internally to improve efficiency. In January of 2026, Think spoke at the World Economic Forum where he boasted about the revolutionary impact of AI.
">> with your data. I could just tell you from at our firm things that would take 12 hours to compute now takes minutes for us processing 14 trillion dollars of other people's money with hundreds of thousands of different um mandates um we could do that instantaneous and we you know that to me if it wasn't for the technology and AI today we would not be able to function to the scale that we're operating. >>"
Black Rockck manages $14 trillion of client assets. Black Rockck's operations are so complicated that it's not possible to manage without AI. Without AI, Black Rockck could not function at its current scale. There are a few problems within narrative. Generative AI only became widely available after Chat GBT was released in 2022. Before then, Black Rockck was not using Generative AI to any significant degree. Back in 2021, Black Rockck already had $10 trillion of assets under management. That's within the same order of magnitude of what they manage now. They were able to function just fine in 2021 without AI. Even after Black Rockck started implementing AI, they continued to hire more employees. Again, no sign of any AIdriven job losses.
When you use LLMs, they can feel like they are very capable. They feel like maybe they should be able to replace at least some human workers. But when left on its own, AI is surprisingly incompetent.
In October of 2025, an organization called the Center for AI Safety published a study called Remote Labor Index, measuring AI automation of remote work. The study was funded by Scale AI. One of the co-authors is Alexander Wang, who is a co-founder of Scale AI and is currently the chief AI officer of Meta. They went on to freelancing platforms such as Upwork and Fiverr. They found freelancing jobs that in theory an AI should be able to complete. For example, build a web page for data visualization, create a 3D rendering of a product, or animate a short video. The average cost for these freelancing jobs was $632.
They tested many different AI models to complete these freelancing tasks. The results were atrocious. The best performing AI agent was Manis. Manis creates AI agents powered by multiple frontier LLM. Even it only had a 2.5% success rate. They gave Manis hundreds of freelancing tasks. Only 2.5% of the time did it complete the job acceptably. Success is defined as creating an output that a reasonable buyer on Upwork or Fiverr would accept. All the other LLM did even worse.
In 18% of cases, the AI agent delivered corrupted files, meaning you couldn't even open the file. In 36% of cases, the project was incomplete. In 46% of cases, the project was completed, but the quality was so bad that a reasonable buyer would not accept delivery of it. In 15% of cases, there were inconsistencies. That means the AI did create a usable output, but it's different from what you asked for. In less than 2% of the cases did the AI complete the job successfully.
The Center for AI Safety maintains a website called remote labor.ai. Every time a new AI model is released, they conduct the test with the new model. Subsequent to them releasing their original report, OpenAI and Anthropic have both released new models. As of the time of recording this video, the best performing model is Claude Opus 4.5. It has a success rate of 3.75%.
This study was funded by Scale AI. Scale AI is a company that sells training data to LLM developers. They have every incentive to exaggerate the capabilities of AI. I give a lot of credit to Scale AI and Alexander Wang for having the integrity to publish this research which exposes the limitations of current AI technology. Daario Amade goes on TV and tries to scaremonger, saying that 50% of entry-level jobs will be replaced by AI within 1 to 5 years. In reality, his best model can't even complete 4% of freelancing gigs. If you can't even do a Fiverr gig, you definitely can't do a full-time job.
To finish off this video, we need to debunk one of the most outrageous recent cases of AI scaremongering. On February 10th, 2026, a guy called Matt Schumer made a lengthy post on Twitter titled, "Something big is happening." In the essay, he claims that AI has reached an inflection point and will imminently take everybody's jobs. Schumer's post went viral, getting over 80 million views on Twitter. It was quickly picked up by the mainstream media. The entire essay was republished in Fortune magazine. Schumer was invited to give interviews on mainstream news outlets, including CNN and The Information.
Once we look at Schumer's resume, you'll see why it's outrageous that this man is given any airtime at all. Matt Schumer is the founder and CEO of an AI startup called Hyperite. It is an AI based text editor. It is powered by a version of Meta's Llama open-source model. In September of 2024, Schumer claimed to have created his own open- source LLM called Reflection 70B. He started with Meta's open source Llama 3.1 model. He used an advanced technique called reflection tuning which enables LMS to fix themselves. This drastically improved his capabilities. Schumer claimed to have tested his new LLM on various benchmarks. On all the benchmarks, it performed better than Chat GPT. On most of the benchmarks, it performed better than Claude. If true, this would be a great achievement. Open AAI and Anthropic spent billions of dollars to develop their Frontier AI models. Matt Schumer, with a tiny fraction of that budget, created a new open- source LLM that performs even better.
There's only one problem. It was all fake. Schumer made Reflection 70B open- source so anybody could test it. People soon found out that it was actually just a rapper that used Claude's API on the back end. People would give it prompts and would give the exact same output as Claude Sonet 3.5. Schumer set up Reflection 70B so that it wasn't allowed to say the word Claude. He didn't want the LM to call itself Claude. If you asked it to say the word Claude, it is programmed to only return blank text. Matt Schumer lied to the public. He said he created a groundbreaking LLM when in reality he created nothing. He just used the Claw API. He went through great lengths to hide this, but he was exposed nonetheless. The fact that the mainstream media takes this guy seriously is a testament to how gullible they are.
You can read Matt Schumer's viral essay if you want. It's on his Twitter account, but it's extremely long and mostly not worth reading. 90% of it is hyperbolic assertions with zero citations or evidence. There's only one part where he makes specific claims. He says he started using Chad GBT 5.3 Codeex. That's OpenAI's latest coding LLM. He says he made an app with Codeex. The app can even open itself and click buttons. Schumer does not explain what this app does. He does not provide any evidence that this app even exists. We're just supposed to take his word for it.
All right, guys. That wraps it up for this video. What do you think about AI taking jobs? Let us know in the comments section below. As always, thank you so much for watching and we'll see you in the next one. Wall Street Millennial signing