Why are more and more companies finding that AI cannot replace human labor?

In recent years, alongside the rapid advancement of artificial intelligence, AI has quickly become widespread in our daily lives, to the point that many companies are even eager to lay off staff as soon as possible, using AI to cut costs, boost efficiency, and achieve all-round profit growth. However, recently some media outlets have raised a different voice: certain companies have suddenly realized they might have downsized too early. Why are more and more companies discovering that AI cannot replace human labor?

1. Regretting downsizing too early?

According to reports from CLS, some experts and critics once predicted that artificial intelligence would deal a severe blow to the global labor market, a scenario that seemed to be corroborated by layoff plans in the tech industry and some manufacturing firms. But new surveys have found that a number of companies now regret their previous layoffs motivated by artificial intelligence.

A report from Intuition Labs notes that focusing budgets solely on replacing humans with technology, without investing in training or skill development, will leave teams unable to use artificial intelligence effectively. Many companies that aggressively pushed automation later regretted their layoffs, because the employees they let go were precisely the key personnel responsible for overseeing AI systems.

Additionally, per a report from Orgvue, 39% of business leaders carried out layoffs due to AI deployment. Yet 55% of those individuals admitted they made the wrong decision when it came to the layoffs.

Recently, U.S. automotive manufacturing giant Ford announced it would rehire hundreds of experienced engineers to resolve quality issues that automated systems cannot handle. Charles Poon, Ford's Vice President of Hardware Engineering, stated that artificial intelligence is a fantastic tool, but its effectiveness depends on the quality of the information you use to train it.

Beyond Ford, Commonwealth Bank of Australia and software giant IBM have also encountered poor efficiency problems after implementing automation. Last year, Commonwealth Bank of Australia laid off more than 40 customer service staff and replaced them with AI voice bots. However, the AI system became overloaded, causing the customer service system to crash and a backlog of incoming calls, forcing the bank to reverse its layoff decision.

Similarly, even Elon Musk, known as the "Iron Man of Silicon Valley," recently announced restrictions on AI usage. According to reporting from U.S. tech media The Information, starting July 6th, Tesla set a $200 weekly cap on AI spending per employee. Any amount exceeding that limit requires separate approval, and the company no longer treats AI as its primary productivity booster.

2. Why are more and more companies realizing AI cannot replace human labor?

Over the past two years, the AI wave has swept through the business world, with many companies seemingly grabbing it as a lifeline to "cut costs and boost efficiency," hastily signing off on layoff lists and treating AI as a universal master key capable of replacing everything. But now, more and more enterprises are coming to their senses and realizing their earlier layoff decisions were far too rushed. How should we view this situation?

First of all, AI is never a perpetual motion machine that runs the moment you plug it in. Many companies' understanding of AI is stuck in the idealized scenarios promoted by internet hype, believing that as long as they connect to a large language model and launch an intelligent system, they can automatically complete work and replace massive numbers of employees — that AI is a perpetual motion machine that runs efficiently nonstop once powered on. But real-world industrial implementation is completely different from what public narratives portray. No enterprise-grade AI application can be deployed once and then used long-term; it heavily relies on continuous data feeding, refined prompt tuning, and iterative correction of massive erroneous outputs. Without ongoing human intervention and optimization, the output accuracy of AI models will keep declining, their adaptability will worsen, and they will eventually become useless decorations.

Many companies previously only saw the obvious cost savings from AI replacing human workers, but completely ignored the hidden labor costs of AI operation and maintenance. Simply put, AI replaces basic execution-level manpower, but creates a whole new set of human roles for data labeling, model tuning, content review, scenario adaptation, and error correction. Many companies blindly laid off frontline business staff and basic operations personnel before realizing that the business experience, scenario data, and problem-solving methods accumulated daily by those employees were the core support for AI to function properly. For example, AI can indeed replace most of a customer service representative's work, but once a company fully replaces its customer service team with AI, it will discover that the real value of customer service — the ability to solve complex problems, the empathy and skill to parse customers' weird and varied expressions to identify their actual needs — is something AI lacks. When human staff are cut, the supporting business data, scenario awareness, and error-correction capabilities disappear too. Not only can AI not work efficiently, it frequently outputs incorrect content that drags down overall business productivity. This is the most intuitive experience for many companies right now: relying purely on AI without human involvement cannot cut costs, but instead leads to large hidden expenses that make the effort not worth the return.

Secondly, companies that implement large-scale AI adoption easily fall into the "factor mismatch" trap. In any production system, all elements must align to generate value. The current reality is that the quality of AI output is highly dependent on human support, especially reviewers with professional judgment capabilities. Many companies laid off their veteran employees who understood the business and had rich experience, leaving behind only a group of operators staring blankly at screens, or management teams that cannot even understand what AI is generating. At that point, no one can verify the quality of AI outputs, and error rates skyrocket.

Worse still, AI is essentially a probability-based generative model that can only solve problems within predefined frameworks. It cannot proactively break established rules, let alone drive market innovation — and this is arguably the biggest difference between AI and humans. What drives business competition? Very often, it's that little bit of "unreasonable" innovation and intuition, which economist Joseph Schumpeter famously defined as "disruptive innovation." This kind of disruptive innovation is precisely what AI cannot provide. The end result is that AI-powered workflows, lacking human judgment and creativity, end up needing humans to step back in, which is an enormous waste of resources.

Thirdly, non-standardized business operations have inherent compatibility issues with AI. The complexity of real business scenarios far exceeds the design boundaries of algorithms. Many business owners assume their operations are standardized, but they are not. The real world is full of gray areas and non-standardized cases. In simple scenarios, like generating a summary or translating a document, AI performs impressively. But once deployed in complex enterprise-level applications, AI immediately struggles to adapt. For instance, when facing a tricky customer complaint or a scenario requiring highly nuanced emotional communication, AI's rigid logic falls far short.

AI lacks empathy; it cannot grasp implied meanings, let alone understand the underlying grievances behind a customer's anger. In such situations, an experienced human customer service representative might calm the situation with a few well-chosen words, but AI would just mechanically repeat policy clauses and drive the customer away. I have personally experienced this frustration with AI customer service at multiple internet companies and commercial banks recently, where AI just loops through repetitive, useless lines that leave you speechless. This sense of powerlessness in complex scenarios has made many companies realize that so-called "standardization" is nothing more than a theoretical utopia. The ability to handle "exceptional cases" in the real world is a company's core competitive advantage — and that is exactly AI's biggest weakness.

Fourthly, at the current stage, AI can undoubtedly greatly improve efficiency — that much is beyond doubt. But expecting it to fully replace human labor is, at least for the foreseeable future, a costly fantasy. Especially in recent times, everyone has noticed that the cost of invoking large language models, namely token fees, has remained persistently high. When you add up API call costs, computing power rental fees, data governance expenses, and the cost of new engineering teams required to maintain AI systems, you will find that using AI to fully replace an average employee is not cost-effective, and may even be more expensive. It's like using a sophisticated, expensive drone to replace a farm worker picking tea leaves with a basket — the math simply doesn't add up.

Therefore, AI's optimal positioning is as an efficiency tool that empowers humans, not a complete replacement. Its best role is to act as a "Swiss Army knife" in employees' hands, freeing them from tedious, repetitive, low-value information retrieval and organization work, so they can focus more on high-level tasks that require judgment, creativity, and emotional connection. Empowering your employees with AI to let an average worker deliver the performance of a core team member is far wiser and more effective than crude, hasty layoffs. After all, the value of a tool lies in being used by humans, not replacing the humans who use it.

Thus, technological replacement of human labor is never a straightforward, linear process that happens overnight. Every technological revolution brings growing pains, but those pains are usually followed by a restructuring of production relations. When facing the AI wave, instead of living in fear and blindly testing in the old world, companies should proactively embrace the new world and re-examine the boundary between humans and machines.

At the end of the day, it is never specific job positions that get eliminated, but the standardized work capabilities that can be easily distilled into automation. As long as humans retain the ability to define problems, empathize with others, and break through established rules, AI will always remain a tool — and humans will always be the masters who wield that tool.