Technological Waves and the Panic Cycle: Why Does the "Occupation Extinction Theory" Keep Resurfacing?
Every time a new model is released, people's anxiety routinely escalates. Recently, the release of Gemini 3, with its powerful front-end generation capabilities, multimodal understanding, and logical reasoning, has basically caught up with or even surpassed ChatGPT 5 in model capabilities. Consequently, self-media outlets have started shouting slogans like "Front-end is dead" and "Designers are dead." It seems that with each technological iteration, society must rewrite its entire list of occupations.
However, looking back at history, this narrative is not new: when the automatic loom appeared, textile workers feared their income would be taken away; when office software became widespread, secretarial and accounting positions were predicted to be drastically reduced or even disappear.
But the reality is that occupations did not collectively vanish as expected; instead, they were restructured: the invention of the loom made cloth cheap, leading to a surge in the number of consumers and consumption frequency, which in turn increased the demand for textile workers. Office software made starting a business easier, leading to a proliferation of startups, which simultaneously drove the expansion of the accounting industry.
One reason the current wave of AI has sparked intense panic is that the speed at which models generate content far exceeds any previous technology. Another reason is that it directly impacts the massive white-collar workforce. However, judging the substitution effect solely based on speed, while ignoring the judgment, responsibility, and interpersonal components of labor, and even overlooking the business model restructuring and market expansion brought about by technological change, inevitably leads to the over-extrapolation of "technological alarmism."
The "death lists" of the past three years reflect more emotion than factual reality. Media and self-media gain traffic from such narratives, while people gain anxiety. What truly changes is not the occupation itself, but the way people accomplish tasks.
From Traffic to Reality: Where Are the Real Changes?
In an attention economy driven by algorithms, panic itself is a highly efficient asset.
Self-media platforms have thus formed a mature narrative and monetization structure: using extreme narratives to create impact, establishing credibility through survivor bias, and ultimately leading to quantifiable monetization paths. This mechanism is not unique to the AI field but is a natural product of the structure of contemporary platform algorithms. AI has become the best subject simply because it more easily triggers structural anxiety among the middle class.
Economic research on automation has long pointed out that technology replaces tasks, not occupations themselves. Within a single job, several repeatable and standardizable parts will be automated away, while parts requiring judgment, coordination, responsibility, and aesthetics will be strengthened. Whether an occupation can survive depends on whether it still contains elements that are difficult to automate.
The following changes in three industries confirm this logic:
- Concept Art and Design: Mass-produced parts are replaced, premiums for aesthetics and complex creation rise.
AI tools have significantly eroded the demand for low-end illustrations, causing the mid-to-low-end outsourcing market to shrink. However, high-end design, which requires style judgment, scene construction, and sustained aesthetic control, has seen its demand and prices remain strong.
According to trend reports from major freelance platforms like Upwork and Fiverr, and industry analysis from the Association of Japanese Animations (AJA), the volume and price of simple, mass-produced outsourcing orders have experienced a structural collapse over the past year. However, during the same period, transaction volumes for high-end projects involving brand visuals and world-building have remained stable and even recorded moderate growth in some niche areas.
This clearly shows that what is disappearing is not the occupation, but its most replaceable layer of "mass-produced labor." Aesthetics, style, and compositional judgment remain scarce resources.
- Software Industry: Junior development decreases, but demand for systems engineering rises.
AI can indeed generate a large amount of basic code, but system stability, architectural design, and risk control cannot be fully entrusted to models.
The software industry is experiencing structural differentiation. Junior development positions face pressure, while engineers capable of managing, reviewing, and integrating AI outputs have become more expensive.
Data from major recruitment platforms (such as LinkedIn and Indeed) depicts this structural differentiation trend. Although demand for junior developer positions fluctuates, recruitment for senior and architect-level positions has been continuously growing, becoming the dominant force in market demand.
A report from GitHub Copilot further explains the underlying reason: although AI assists with a large amount of programming work, due to increased difficulty in system integration and risk control, the number of Pull Requests has actually increased. In other words, AI hasn't reduced "engineering"; it has only reduced "writing code," ultimately requiring more people to manage the complexity introduced by AI (commonly known as "technical debt"). AI replaces repetitive labor, not engineering capability itself.
- Translation Industry: Basic translation is automated, high-end translation shifts towards responsibility and guarantee.
Instruction manuals and daily communication are highly automated, but legal contracts, diplomatic language, and literary translation depend on authorial intent and legal responsibility. The core value of high-end translation has instead been upgraded to "quality assurance."
Industry data from Proz.com and the American Translators Association (ATA) shows that prices for ordinary text translation (manuals, general business materials) have experienced significant declines. However, prices for translations involving liability, legal validity, and stylistic consistency—high-risk or high-context-dependent types—have remained stable or even seen moderate growth.
Although the usage of mainstream machine translation tools like DeepL and Google Translate has increased significantly, this only affects "dictionary-lookup labor." In scenarios involving significant risk, publishers and law firms generally insist on a collaborative model of "AI first draft + human final review." That is, AI automates language transcription but cannot automate contextual judgment. The group most impacted is not the industry itself, but those who relied on repetitive labor.
The Boundaries of Automation: What is the Core Competitiveness of Humans?
The speed at which models generate content is indeed astonishing, but the clearer we understand their boundaries, the better we can see the true advantages of humans. Automation has advanced steadily, yet it has always been blocked by three "insurmountable thresholds," and these three thresholds precisely constitute the core competitiveness of humans.
The first threshold is responsibility: In any field involving risk where someone must bear the consequences, the final decision can only ever be made by a human. Financial regulatory agencies explicitly emphasize in documents: AI can assist, but cannot replace humans in assuming ultimate responsibility. Regulations in the medical field are even stricter, requiring all AI outputs to be reviewed by qualified professionals. This means that humans must be the ultimate bearers of risk. Models can serve as advisors, but they cannot be decision-makers.
The blame for AI's mistakes always falls on humans.
The second threshold is context: The real world is not standardized input; it is full of ambiguity, subtext, conflicts of interest, and unspoken rules. AI may appear flawless in text, but once it enters "gray scenarios," its error rate skyrockets. Research has found that models struggle most with complex "gray scenarios" requiring deep situational understanding and interest coordination, which are precisely the core work of many occupations.
"Have you eaten?" is not really asking if you have eaten.
The third threshold is relationships: The essence of many occupations is not "providing information" but "providing relationships." Sales, consulting, medical services, psychological support—they rely on trust, emotional coordination, and subtle interpersonal cues. A model can be very intelligent, but it cannot provide the necessary psychological support and sense of security.
AI cannot provide warmth and hugs.
Understanding these three thresholds makes it clear that AI changes the structure of tasks, not the labor system itself. Machines can replace repeatable actions, but not the critical responsibilities humans bear in an uncertain world. Therefore, the core competitiveness of humans in the future labor system is a direct response to these three "insurmountable thresholds." Future competitiveness will increasingly concentrate on abilities that cannot be written into algorithms:
- Responsibility threshold corresponds to final judgment and accountability: The ability to make final decisions based on risk and ethics and bear irreversible consequences. This is the last line of defense in turning data into action.
- Context analysis threshold corresponds to complex situation and cross-domain integration ability: The ability to define ambiguous, non-standardized real-world problems, identify subtext and conflicts of interest, and use AI as a tool to integrate solutions across multiple domains and factors.
- Relationship building threshold corresponds to emotional intelligence and interpersonal trust building: The ability to provide empathy, reassurance, and emotional support, and build trust through interaction. This is key to providing core services like security, understanding, and support.
These abilities together form the "foundation of irreplaceability" in the AI era.
The Threat of Technology is Often Overestimated, While Human Potential is Often Underestimated
The emergence of generative AI has indeed changed the task structure of many industries, but "occupational extinction" is more of a narrative than a reality. History has repeatedly proven that technology can change the way we work, and the application of new technologies can generate more demand. Therefore, the real risk is not the capability of the model, but humanity's misinterpretation of technology.
AI will not eliminate people, but people who skillfully use AI will eliminate those who refuse to change. The future belongs to those who can ask good questions, make deep judgments, take on complex responsibilities, and create value on top of tools.
Perhaps the final question should be: Is it AI that eliminates you? Or did you choose to give up on yourself first in the face of technological change?
Moreover, the impact of this wave of technological change extends far beyond individuals; it will also reshape or even eliminate organizational structures themselves. History repeatedly proves that companies clinging to old models at technological turning points are often more easily eliminated than individuals. If a company cannot reshape its processes, culture, and decision-making methods, even if it is large, it may quickly lose its competitiveness.
In other words, the risk in the AI era is never simply "individual versus machine," but a systemic challenge: Can individuals and companies jointly adapt to new modes of production?
The ultimate competition is not between humans and machines, but evolves into a life-or-death race between evolving organizations and those clinging to old habits. Only those organizations brave enough to restructure processes and embrace uncertainty can truly build a fortress for the future amidst the wave of automation.