The AI Replacement Wave: Can Switching to a Trade Really Save You?

In discussions about AI's impact on employment, one piece of advice is becoming increasingly popular: "Abandon office work" and "go learn a trade." The logic seems sound—white-collar jobs are highly digitized and process-driven, making them precisely the kind of work that large language models and automation tools can most easily penetrate. Meanwhile, blue-collar skilled trades like plumbing, electrical work, and carpentry require complex hand-eye coordination and on-site judgment, making them difficult for algorithms to replace in the short term.

But a more sobering perspective is gaining attention: once algorithms become more powerful, deployment becomes cheaper, and applications become more widespread, "switching to a trade" may not serve as a long-term safe haven. From the perspective of some corporate executives, eliminating all forms of human employment is the ultimate goal—not just white-collar workers, but all replaceable labor.

This claim deserves serious examination. It contains both reasonable extrapolation of technological trends and a degree of emotionally-driven prediction. We need to distinguish between genuine risks and amplified anxieties.

White-Collar vs. Blue-Collar: Different AI Replacement Timelines

Why Office Work Gets Hit First

The jobs currently facing the greatest AI replacement pressure are indeed knowledge-based, process-driven positions. Copywriting, junior programming, data organization, customer service, legal document drafting, and basic financial analysis—these jobs share one thing in common: both inputs and outputs are digital information, requiring no interaction with the physical world. Large language models are naturally adept at processing text, code, and structured data, so the conversion of these roles into efficiency tools—or even their partial replacement—comes fastest and at the lowest cost.

The reason LLMs can penetrate these fields so rapidly is directly related to their underlying technical architecture. Models like GPT-4 and Claude are based on the Transformer architecture, using attention mechanisms to capture long-range semantic dependencies in text, with parameter scales reaching hundreds of billions. After pre-training on massive internet text corpora, they acquire broad world knowledge and reasoning patterns. This means that any workflow where inputs can be described in text and outputs expressed as text or code can theoretically be approximately executed by models. More critically, there's the deployment cost—a single API call might cost just a few cents, while a human employee completing the same task requires an hourly wage of tens of dollars. This order-of-magnitude cost difference leaves enterprises almost no room for hesitation in their ROI calculations.

How Long Can the "Physical Moat" of Skilled Trades Hold?

By comparison, blue-collar technical trades currently do have higher barriers to replacement. An electrician troubleshooting an unfamiliar, aging electrical panel involves spatial perception, real-time decision-making, and fine motor operations that remain challenging for robotics. This is the core basis for the "learn a trade" advice: the complexity of the physical world and unstructured environments form a natural barrier.

But this barrier isn't permanent. As embodied AI and robotics advance, combined with declining hardware costs, the automation of physical labor is a matter of when, not if. Embodied AI refers to deploying AI algorithms in robotic systems with physical bodies, enabling them to perceive, understand, and manipulate the real world. Representative advances in this field include Google DeepMind's RT-2 model combining vision-language models with robot control, and humanoid robot projects like Figure AI and Tesla Optimus attempting to achieve general physical manipulation capabilities. However, significant technical bottlenecks remain—long-tail scenarios in real environments such as lighting changes, object deformation, and unexpected obstacles are extremely difficult to enumerate exhaustively; force feedback control precision for fine manipulation still falls far short of the human hand; and the hardware cost of a single general-purpose robot currently remains in the tens to hundreds of thousands of dollars range. Yet Moore's Law-style cost reduction trends and the proliferation of open-source robot learning frameworks are continuously narrowing this gap.

Historically, industrial robots have already replaced a massive number of manufacturing jobs; today's progress simply extends this replacement to more non-standardized scenarios. Looking back at this history, industrial robots' replacement of manufacturing began in the 1960s when General Motors introduced the Unimate robotic arm, subsequently progressing through three phases: replacement of dangerous repetitive work like welding and painting, automation of electronics assembly and precision machining, and flexible manufacturing represented by collaborative robots (cobots) from the 2010s onward. According to the International Federation of Robotics, the global installed base of industrial robots has exceeded 3.8 million units. The key historical lesson: every wave of automation begins with the most standardized, most repetitive segments, gradually expanding toward more complex non-standard scenarios—blue-collar service industries currently sit at the frontier of this expansion path.

Does the Claim "The Ultimate Goal Is to Eliminate Employment" Hold Up?

The True Motivations of Corporate Executives

The most controversial claim is that "executives see this as the endpoint for all human employment." This requires a dialectical perspective. From the logic of corporate finance, labor costs are often the largest variable expenditure, and any technology that can reduce this cost will be actively adopted—this is a basic principle of capital operation, requiring no conspiracy theory to explain.

But "eliminating all employment" isn't an explicit executive goal—it's the natural corollary of cost optimization. Enterprises pursue profit; reducing headcount is merely a means. When the marginal cost of automation falls below that of human labor, replacement will occur; otherwise, it won't. Therefore, a more accurate framing is: within the bounds of what technology and cost allow, enterprises will continuously substitute capital for labor.

The Overlooked Structural Contradiction

Here lies a classic economic paradox: if all enterprises massively cut employment, who will consume the products and services they produce? Mass unemployment would destroy purchasing power, ultimately harming the enterprises themselves. This contradiction means that a purely "zero employment" society cannot operate stably within existing economic structures, and would likely force social-level adjustments such as Universal Basic Income (UBI), reduced working hours, and wealth redistribution.

UBI, as a potential solution to this structural contradiction, has moved from thought experiment to empirical testing. The concept traces back to Thomas More's Utopia, with its modern version grounded academically by Friedman's "negative income tax" and Van Parijs's theory of justice. Finland, Kenya, Stockton (California), and other locations have completed or are conducting empirical studies, with preliminary results showing UBI can reduce participants' anxiety levels and increase entrepreneurial willingness, though its effect on overall labor participation rates remains debated. In the scenario of large-scale AI labor replacement, UBI is seen as a potential safety valve for maintaining the social consumption cycle, but its funding sources (automation taxes, carbon taxes, data taxes, etc.) and inflationary effects remain unsolved problems requiring more large-scale, long-term experiments to verify feasibility.

Personal Strategies for the AI Era: From "Picking the Right Track" to "Maintaining Adaptability"

Why a Single Risk-Avoidance Strategy Will Fail

The real insight may be this: trying to avoid risk once and for all by "picking the right profession that won't be replaced" is itself an outdated way of thinking. In an era of accelerating technological change, any static career choice could be redefined within a few years. Today's safe trade could be tomorrow's replacement target.

Four Capabilities More Important Than Choosing the Right Track

Rather than betting on a specific position, build a combination of capabilities that cannot be replaced by any single technology:

Human-AI collaboration ability: Learn to harness AI tools rather than compete with them; become an efficiency multiplier. Human-AI collaboration isn't simply "using AI tools"—it's a paradigm for redefining workflows. Humans are responsible for defining problem frameworks, providing value judgments, and quality control, while AI handles information retrieval, draft generation, and solution enumeration. Typical examples include lawyers using AI for case research before constructing their own argumentation strategies, designers using generative AI for rapid prototyping followed by manual refinement, and doctors leveraging AI-assisted diagnostic systems to make final clinical decisions. McKinsey research shows that human-AI collaboration can boost knowledge worker productivity by 40-80%, but this requires workers to develop prompt engineering thinking, output verification skills, and habits of critically evaluating AI outputs—which itself is a new professional skill requiring deliberate cultivation.
Cross-domain integration ability: Machines excel at single tasks; humans' advantage lies in synthesizing judgment across contexts.
High-context, high-trust work ability: Roles involving complex interpersonal relationships, responsibility-bearing, and on-site decision-making are harder to replace.
The meta-ability of continuous learning: Adapting to change itself is more valuable than mastering any specific skill.

Conclusion: Rational Judgment Beyond Anxiety

"Switching to a trade to escape AI" is a comforting narrative, but the truth is more complex: there are no absolutely safe positions, only relative time differences and varying adaptive capabilities.

At the same time, we needn't fall into the fatalism of "humanity will ultimately be rendered completely obsolete." Technological replacement is constrained by cost, social tolerance, and economic cycles. Every wave of automation in history has been accompanied by the creation of new jobs and adjustments to social institutions. The real challenge isn't the technology itself, but whether society can build supporting distribution mechanisms and retraining systems in time, so that the pain of transition doesn't crush the most vulnerable groups.

For individuals, the most rational strategy isn't finding a permanent safe harbor, but making yourself someone who can continuously adapt amid uncertainty. This, perhaps, is the most important "craft" of our era.