You have seen the headlines: Story after story is predicting doom for the job market.
Some are taking Anthropic CEO Dario Amodei’s commentary that “AI” will eliminate half of all entry level white collar jobs within five years as gospel, and opining on what will happen when workers are replaced. In the most egregious prediction I’ve seen, Roman Yampolskiy, an associate professor at the University of Louisville and self-described “AI expert”, predicts that 99% of work will be replaced by “AI” by 2030.
“That is not even a question, if it’s ‘possible,’” Yampolskiy said recently on the Diary of a CEO podcast. “It’s like, how soon before you’re fired?”
We are starting to see trends in the opposite direction ocurring.
We are starting to see trends in the opposite direction occurring: Large language model performance may be plateauing or going backward, and businesses are realizing that these algorithms will require more human workers, not less. As of August 2025, Microsoft had sold 8 million Microsoft 365 Copilot subscriptions. That sounds impressive until you consider that more than 440 million people subscribe to Office 365, equating to a conversion rate of less than 2% to existing customers.
(Since I am now on the record here, I will check my predictions against Dario Amodei and Yampolskiy in 2030.)
I make a practice of putting “AI” in quotation marks, because it is a marketing term, not a technical term: The technologies under its umbrella are neither artificial nor intelligent.
When we put buzzwords aside, what we are discussing are algorithms and automation. When most people use the term “AI”, they are not specific, and most often mean large language models (LLMs). LLMs are just one kind of predictive algorithm that is currently seeing mass publicity. While it is portrayed that these technology advances appeared almost overnight in 2023, the reality is they have progressed gradually over decades. I disagree with the current sentiment that “AI will take all the jobs,” because of the rich history of technical advances that provide us a roadmap to examine.
A review of tech innovation history
In 1840, nearly 70% of the American workforce worked in agriculture. Since then, we have seen a steady decline in the number of farms and their employment, yet we have had a continual increase in the number of jobs in the US over almost two centuries.
The dramatic shift away from agricultural employment, which shrank from 70% to 5% by 1970, did not decimate the American workforce, because new industries and occupations absorbed the displaced workers.
If we narrow our scope to the past 75 years and hired farmworkers, rather than agriculture as a whole, the number of workers continues to diminish while the amount of food produced continues to increase. This is due to technological automation of manual farm labor. The machines may have displaced manual labor, but they created a net-gain in overall jobs. It was a shift in the labor market, rather than a decimation.
The “AI” hype wave is the largest I have seen in my four decades in technology. The introduction of the personal computer itself wasn’t hyped nearly as much, but if there is any historical tech disruptor that would be actually worthy of the current wave, surely it would be the 1980s–1990s introduction of the PC into business.
That did eliminate a lot of jobs…at first. The introduction of the PC eliminated 3.5 million jobs, according to a report from global consulting firm McKinsey, but then created over 19 million new positions.
While jobs for typists, secretaries, and paper filers largely disappeared, entire new fields were created. Data analysts, video game designers, software developers, UX architects, influencers and many other job categories now exist because of the introduction of the PC.
Two decades ago, many analysts and media headlines were declaring that “technology” would eliminate a large percentage of jobs in the health care sector. Instead, health care employment has doubled. The current predictions sound very familiar to what we were hearing just a few decades ago, except “technology” was the inexact term being used instead of “AI”.
Is ‘AI’ just an excuse for bad hiring policy?
A majority of the current over-valuation of the market stems from the belief that “AI” will make jobs redundant. It feels gross to see market gurus downright gleeful about people losing their jobs. Let’s take a look at Marc Benioff of Salesforce as an example.
Benioff claimed that “AI” powered systems allowed him to cut one of his teams from approximately 9,000 employees to 5,000 employees, a nearly 45% reduction. (The first eight months of 2025, during which Benioff eliminated 10,000 of his employees’ jobs ostensibly because of “AI”, he described as “eight of the most exciting months of my career.”)
Many media outlets ran headlines citing Benioff’s declaration as definitive proof of the effect of “AI” and what it portends for the future. Not a single article I found talked about the previous five years of Salesforce hiring, which would seem salient and obvious to investigate before parroting such bold declarations.
Salesforce is a good case study. During the COVID pandemic, I frequently commented that big tech was “hiring stupid,” and there would be a correction. By the end of 2022, Salesforce had 30% more employees than in 2021, ballooning to over 70,000 staff. That means they hired over 20,000 people in a year.
Salesforce soon noticed that the 20,000 people they had hired during their bender were far less productive. That would have been an easy prediction to make, because vetting during the interview process is abandoned when hiring that many people. Additionally, how can a company successfully onboard 20,000 people in one year, with an existing staff of 50,000? It was gross incompetence to hire so recklessly. Benioff later admitted as much.
So what is more likely? Is Salesforce actually replacing their staff with some “AI” magic wand, or is a CEO covering for a bad policy, a reckless hiring binge, while simultaneously getting the benefit of boosting his company’s stock price by inserting “AI” into every other sentence he utters?
What is the real cause for the anemic job market?
The past few years were riddled with bad decisions, leading to this difficult job market so many of our fellow citizens are experiencing. Some incredibly skilled former colleagues of mine have been on the open market for months looking for work.
If you look at some occurrences of the past decade, this could be seen coming:
- A huge tax cut, mainly for the wealthy, was passed during a roaring economy. Most economists worth their salt agree: Tax cuts are best used as a stimulant during tough times, not when the economy is good. This created a sugar rush, and the subsequent sugar crash.
- When you take the tax cut, and synergize it with a haphazard reaction to a pandemic, it multiplies the economic fallout several years down the road. With our society becoming fully online overnight in March 2020, the tech sector went into hiring overload; it wasn’t just Salesforce. A correction was inevitable.
- The United States is adrift on an ocean of economic uncertainty, driven in no small part to tariffs being applied by whim, and policy applied by social media post. Nick Baker, Managing Director of Trade and Customs at financial risk advisory firm Kroll, says “it’s maximum chaos.” CEOs are dialing back plans to hire or expand. It is difficult to plan when you don’t have a good idea what might change in the next 12 months, let alone when you don’t know what might change in the next 12 hours.
The “maximum chaos” we’re seeing in the United States doesn’t allow employers to plan. Enough risk is created by just existing in the current environment that very few employers have the bandwidth to take on more. This leads to very little real innovation, expansion or new hiring.
Therein lies the bubble creating jobs
The hyped-up pitch that “AI” companies are making to investors is that the tech will replace a huge amount of jobs, with the firing company and the “AI” company splitting what used to be a salary.
Even if this was true, it would create a huge problem: What do we do when there’s a huge unemployment crisis? Fortunately, that isn’t going to happen. I have yet to see “AI” that can do an entire job function well. LLMs can sometimes help us do our jobs more efficiently, but there have been many, many tech innovations over history that have done that. Additionally, studies are showing an interesting cognitive paradox: “AI” coding tools actually make developers slower, even though they think they’re moving faster.
While writing this article, I decided to enlist OpenAI’s flagship model at the time, ChatGPT 5.1, with two fairly straightforward tech issues. I have been underwhelmed with LLM improvements over the past several years, since ChatGPT 4’s release, and wanted a few examples from real-life to include here.
The first problem I posed with this prompt: “Can I set up a Meta Quest 2 with a point-to-point WiFi 6 link to my Windows 11 PC for streaming with SteamVR Link instead of running it over my WiFi router for greater speed?”
The LLM confirmed what I had hoped: Yes, I can use two WiFi adapters on my home PC. One would connect to my WiFi router, and the other would be dedicated for the Quest 2 to connect to my PC directly. Over the course of the following week, I ordered four different USB WiFi adapters that were “guaranteed to work” by ChatGPT 5.1. None of them have. One did what I needed successfully, but was only a USB 2.0 interface, which is slower than WiFi 6, eliminating any advantage. All three of the others were incompatible as hotspots when used with the already existing device to connect to the WiFi router.
After going around in circles for hours over several days, and avoiding some very bad advice that could have completely broken my networking setup, I resolved to figure out the problem the old-fashioned way. I haven’t gotten it working yet, but I will return to the tried-and-true methodology of talking to some colleagues with expertise in networking, and using the DuckDuckGo search engine. I wasted an hour returning the four different USB WiFi adapters at the UPS Store yesterday, after ChatGPT 5.1 promised they were “guaranteed to work.”
The second issue I presented ChatGPT 5.1 was to uninstall a program.
I had installed NVIDIA’s “Chat with RTX” program as an experiment with a local LLM on my home PC. I was curious how a local model would perform in contrast to the web-based interfaces that ran in the cloud. After several hours of being given commands to run that didn’t work, and going around in circles., I found the solution to uninstalling the program myself.
I used WinDirStat to find the directory it had actually installed to, since it had a huge model as part of the program payload. Then I used DuckDuckGo to find out how to remove a lingering start menu entry. After pointing this out to ChatGPT 5.1, it predictably responded in obsequious mode.
I use these two examples to underline the extreme limitations of LLMs. OpenAI lost $5 billion last year, making $4 billion in sales…while spending $9 billion.
They did this without any regard to consent or copyright, and this is the product we have. It is not a coincidence that LLM performance has plateaued over the past few years, when they ran out of data to grab for their training sets. There is literally not more data for them to take.
The current economic bubble is the gap between the C-suite believing that “AI” can replace workers, and the reality that it cannot, and I am much more worried about the economic fallout from this outrageous, irresponsible hype-cycle than I am about the technology killing the job economy. While I was writing this article, Cory Doctorow published a transcript of a talk he gave which is an essential read, in which he comments:
“This is another key to understanding — and thus deflating — the AI bubble. The AI can’t do your job,” Doctorow said, “but an AI salesman can convince your boss to fire you and replace you with an AI that can’t do your job.”
Following the historical pattern, ‘AI’ is actually creating jobs
Healthcare seemed like a prime candidate to have employment dwindle due to technology automation over the past few decades, but we saw the opposite. The trend in healthcare seems to be repeating itself: More radiologists are being hired, because the new “AI” tools in their toolbox are creating new jobs requiring oversight. The volume of imaging workloads are increasing due to them becoming a more useful tool for early detection.
Financial Times author John Burn-Murdoch writes, “I find this a fascinating demonstration of why even if AI really can do some of the most high-value parts of someone’s job, it doesn’t mean displacement (even of those few tasks, let alone the job as a whole) is inevitable.”
This seems to be an early beacon of hope for hiring to come. If our country can dial back the “maximum chaos” to create more certainty in the business environment, hiring should follow.
Hope is on the horizon
There has been pushback on the trend of replacing junior employees. Matt Garman, the CEO of core “AI” player Amazon Web Services, recently called junior staff replacement, “one of the dumbest things I’ve ever heard.”
Garman continued: “They’re probably the least expensive employees you have. They’re the most leaned into your AI tools. How’s that going to work when you go like 10 years in the future, and you have no one that has built up or learned anything?”
LLMs and other algorithms under the “AI” umbrella are new tools in my work toolbox, but they don’t change the fact that principal staff come from senior staff, and senior staff come from junior staff.
Continuing to hire graduates, teach them best practices, and how to solve interesting problems will lead to far more productivity than LLMs ever will. I like to remind people that the biggest boost to my productivity as a software engineer hasn’t been LLMs, or even software intelligence or autocomplete: It was Windows 98. That was the OS that allowed me to have a second monitor.
There is no doubt that the job market is incredibly tough right now, for a number of reasons. I don’t believe “AI” is to blame; the primary factor is the unpredictability coming from the top levels of the United States government. For those who are currently looking for work, I do sympathize.
The history of technical innovation should offer comfort: Over the past two centuries, every major technical innovation has led to more jobs, and better jobs. There is no reason to believe it will be any different this time.
