The AI Boomerang Is About To Hit Hard
Reality vs speculation.
I, along with many others, have been yelling from the sidelines that AI can’t replace jobs for a while now. They cost too much, are far too inaccurate to be genuinely autonomous (given that they aren’t actually intelligent), and don’t increase productivity in a way that threatens human labour on any meaningful scale. Well, it turns out the people who have been almost pervertedly pushing for AI to eradicate human workers have only just discovered the truth and are now rapidly U-turning on this ‘revolutionary’ technology. Who would have guessed?
There is a lot of evidence out there that the corporate world is rapidly bailing on AI.
Forrester Research found that 55% of employers who laid off workers because of AI now regret that decision. OrgVuefound something almost identical, with 55% of business leaders admitting they made the wrong call eliminating jobs and replacing them with AI. Earlier this year, Gartner predicted that half of all companies that replaced customer service or operational employees with AI will be forced to restaff those roles. Sinch, a cloud communications company, surveyed 2,500 AI decision-makers across multiple industries and countries and found that AI customer communications agents have a 74% rollback or shutdown rate. Customer communications was once considered one of the easiest jobs for AI to replace, yet it is failing catastrophically. Robert Half and Orgvue found that 32% of HR say organisations have already had to rehire positions previously replaced by AI, explicitly because AI couldn’t actually replace these roles.
This has been dubbed the “AI boomerang effect”. This is when companies that once bullishly ploughed headlong into AI layoffs are having to rapidly U-turn, given that their ignorance and unfounded confidence in the technology has been laid bare.
But why is AI failing to replace these jobs? Well, there are actually multiple reasons — namely, cost, hallucinations, and lack of productivity gains.
I have written about the AI cost issue before. In an attempt to stem their spiralling losses, AI companies like OpenAI and Anthropic have switched from using heavily discounted flat-fee subscriptions to token pricing. This way, the amount a customer pays for an AI more closely reflects how much it costs to operate. These token prices are almost certainly still so heavily discounted that OpenAI and Anthropic lose money on them, but this allows them to better control their losses.
However, this change has made many AI tools skyrocket in price. For example, the cost of GitHub’s Copilot-based AI has increased tenfold; Uber has said that AI is more expensive than human workers; Microsoft has cancelled its Claude Code licence because it is too expensive; and one company even racked up a $500 million bill with Anthropic in a single month.
Quite simply, even if AI were able to replace human labour, it would be more expensive than human labour in the overwhelming majority of cases, making AI layoffs a pretty dumb business move. Particularly when these AI companies are still selling their models at a catastrophic loss and will almost certainly have to massively jack up the prices down the road. So, firing cheaper labour is doubly stupid because it makes your business dependent and vulnerable.
But, here is the thing: AI isn’t capable of replacing most jobs because it isn’t actually intelligent; it’s just a statistical model, meaning it makes constant errors, also known as hallucinations.
AI hallucinations prevent these machines from being able to do tasks autonomously. For example, Carnegie Mellon University found that even the best “agentic” AIs fail basic tasks 70% of the time. Another study found that the best current AIs failed 97.5% of realistic real-world freelancing jobs. Both of these studies specifically highlight AI hallucinations as the primary cause of this astonishing failure rate.
You can’t simply ask an AI to complete a task; it requires extensive human oversight to catch and mitigate these errors. In no task is this issue better exposed than in coding. The University of Waterloo found that even the best generative AI coders only have a 75% accuracy rate when tasked with very basic coding tasks. Research from Veracode found that 45% of AI-generated code contained security flaws. A study from Coderabbit found that AI-generated code has 70% more bugs than human-written code. In fact, both Anthropic and METR found that AI coding tools actually reduce productivity, because the time the coder saved by using AI to write the code is less than the time they spent debugging the AI-generated code. In other words, it is often faster to write and debug code by hand than to use AI tools to code and spend hours trying to debug it to an acceptable level.
We often see this problem with AI. It might appear to dramatically increase your productivity for a single task, but once you zoom out and look at your overall productivity, those gains totally disappear. As such, we don’t actually see any AI productivity gains on an economic, organisational, or individual scale. Take this study published in February, which surveyed 6,000 CEOs, CFOs and other C-suite executives across various countries. Nearly 90% of these firms said AI has had no impact on employment or productivity over the past three years. Likewise, the recent Oxford Economicsreport says that if AI were increasing productivity enough to replace jobs, we would see that in economy-wide productivity data, but we don’t; our productivity is actually quite stagnant. In other words, AI does not increase overall productivity in the overwhelming majority of cases.
You might think that with the current colossal investment in AI, these systems will become better and soon solve these problems. But that isn’t the case. Even OpenAI has admitted that they can’t reduce hallucinations from their current level, and the cost of operating these models is only increasing (read more here). There will be tweaks and refinements that make these models less dreadful, such as with Anthropic’s new batch of models, but they will still suffer from these critical flaws.
The entire AI bubble is predicated on the idea that AI will take jobs and generate hundreds of billions of dollars in annual revenue within just a few years. If that doesn’t happen, the AI industry can’t service its debts, and the whole circus collapses.
The fact that the AI boomerang effect exists at all is seriously bad news for the AI bubble. But we have only just begun to witness this U-turn. It takes time for companies to notice that AI is damaging their bottom line. Many companies feel behind on the trend of adopting AI, and so are blindly and rapidly adopting it out of FOMO. When AI companies run out of cash to burn, they will drive up prices, causing more businesses to ditch them. All of these factors suggest that the AI boomerang effect is likely only just getting started and that we will only see this trend strengthen over time. What does that mean for the economy, businesses, or the AI bubble? I don’t know — I don’t have a crystal ball. But what I do know is that this is not good news for the AI industry at all.
Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!