Is the AI replacement trend overestimated? Barclays: Only 20% of core workplace skills can be highly replicated, but the amplification effect has already quietly spread.

The impact of AI on the labor market may be more subtle, but also more profound, than the market narrative suggests. A recent Barclays study shows that only about 20% of core workplace skills can be highly replicated by AI, suggesting that concerns about a large-scale "replacement wave" are overestimated. However, the "amplification effect" of AI on productivity presents a completely different picture—its reach is far broader than automation replacement, penetrating almost every corner of the labor market.

On September 22, Barclays released a report introducing the "AI Automation and Amplification Framework" (Barclays 3A Framework), which uses skills rather than occupations as the core analytical unit to systematically assess the impact of cognitive and physical AI on over 830 occupations. The study found that automation risks are highly concentrated in a few occupational groups. Occupations with high exposure to cognitive AI are mainly concentrated in computer science, engineering, and science positions, while those with high exposure to physical AI are concentrated in construction, manufacturing, installation, maintenance, and repair. Meanwhile, US recruitment data has shown a structural change consistent with AI—since 2022, the percentage of job postings for occupations with the highest automation exposure has fallen from approximately 30% to around 25%.

The rapid expansion of demand for AI skills reveals another underlying logic: in the United States, the United Kingdom, France, and Germany, the proportion of positions requiring AI skills has climbed from approximately 2% in 2019 to 6% to 10% by August 2026. It is noteworthy that the focus of AI skills demand is shifting—in 2019, over 70% of AI-related positions were concentrated in technology and data roles, but by 2026 this proportion has dropped to approximately 50%, while the proportion of management positions has risen from approximately 10% to 25%, indicating that AI is spreading from specialized technical fields to a broader workplace ecosystem.

AI automation may amplify specific skills rather than the entire profession itself.

Traditional AI workforce impact analyses often focus on "occupations," but Barclays argues that an occupation is a collection of tasks, which in turn are collections of skills. Skills are the "atomic units" for understanding the impact of AI. AI is more likely to automate or amplify specific skills—such as information gathering, writing, or data analysis—rather than the occupation itself.

The 3A framework is based on the U.S. Department of Labor database, which covers 104 cross-occupational skills and maps them to over 800 occupations. The framework's innovation lies in three dimensions: First, it simultaneously incorporates cognitive and physical AI, encompassing both digital and physical intelligence; second, it uses skills as the basic unit of analysis, replacing tasks or occupations; and third, it simultaneously captures both the "automation" and "amplification" effects—the former measuring the degree to which AI substitutes for human labor, and the latter measuring the potential of AI to enhance human skill productivity.

The research results show that only about 20% of workplace skills are rated as highly replicable by AI, 40% of skills are relatively resistant to cognitive AI, and the proportion of skills that are more resistant to physical AI is as high as 70%, indicating that the current automation impact of physical AI is still far narrower than that of cognitive AI.

Automation exposure is highly concentrated, and most of the workforce is relatively safe.

Barclays' key finding regarding the distribution of automated exposure is that extremely high exposure levels are concentrated in only a very small number of occupations, with most occupations primarily exhibiting exposure to either cognitive or physical AI, rather than both.

Occupations with the highest exposure to cognitive AI automation include data scientists, statisticians, and actuaries, whose core skills—information retrieval, pattern recognition, document processing, and programmed decision-making—highly overlap with AI capabilities. Occupations with the highest exposure to physical AI, such as agricultural workers, textile machine operators, and structural steelworkers, rely on repetitive manual labor.

In contrast, approximately 120 occupations have relatively low overall exposure to both types of AI automation, covering about 20% of the U.S. workforce. These occupations mainly include teachers, childcare workers, coaches, and some hospitality and entertainment workers—occupations that rely heavily on interpersonal interaction, on-site supervision, or entertainment performances and are difficult for AI to replicate.

Amplification effect: Wider coverage, greater potential benefits

Unlike the centralized distribution of automation, the amplification effect of AI is more widespread and uniform. In Barclays' model, the amplification effect is most pronounced in occupations where the proportions of replicable skills, partially replicable skills, and non-replicable skills are most balanced—AI takes over the replicable portion while enhancing the value and productivity of the remaining human skills.

The professions with the highest cognitive AI amplification effect include management positions such as air traffic controllers, healthcare service managers, and training and development managers. AI can take over administrative document and information processing tasks, while amplifying core human competencies such as leadership, coaching, and personnel management. The physical AI amplification effect is most prominent in various mechanical and technical professions, such as bus and truck mechanics and industrial mechanics. Some routine physical tasks can be automated, while diagnosis, decision-making, and response capabilities still require human intervention.

Barclays points out that the core information of the amplification effect distribution lies in the fact that while AI's automation capabilities are concentrated in specific professions, its productivity-enhancing capabilities cover a very broad range of professions. Doctors, teachers, and engineers, among other diverse professions, can all substantially benefit from AI tools. This characteristic implies that the overall productivity gains from AI may exceed its direct substitution effect in the long run.

Recruitment data confirms structural shifts, with demand for AI skills spreading to management levels.

Labor market data has begun to corroborate this framework. According to Barclays' analysis of LinkUp recruitment data, since 2022, the percentage of job vacancies for occupations with the highest automation exposure has decreased from nearly 30% to about 25%, with the decline in software engineering positions contributing the most; while the percentage of job vacancies for occupations with the lowest automation exposure has risen slightly from about 20% to 21% to 23%.

At the same time, the demand for specialized AI skills is expanding rapidly. According to data from Indeed Hiring Lab and other sources based on broader keyword searches, the proportion of AI-related jobs in the United States, the United Kingdom, France, and Germany has risen from approximately 2% in 2019 to between 6% and 10% in 2026, with the United Kingdom showing the most significant increase.

More structurally significant is the shift in the main drivers of demand. In 2019, technology and data positions accounted for over 70% of AI-related job postings; by 2026, this proportion has dropped to approximately 50%, while the proportion of management positions has risen from approximately 10% to approximately 25%. This trend is evident in the United States, the United Kingdom, France, and Germany, indicating that AI is permeating the entire workplace as a general job skill rather than a proprietary technology.

The Rise of Physics AI: Skills May Become Licensable Intellectual Property

The report further anticipates the new economic models that will emerge from the widespread adoption of physical AI, dividing them into two phases.

In the first phase, humans act as "robot trainers"—because physical AI lacks real-world datasets equivalent to the training corpora for large language models, the industry is relying on human demonstrations to generate training data for robots through methods such as remote control and first-person video capture. Companies like Figure's Index and Scale AI are already building this crowdsourcing ecosystem.

In the second phase, the focus will shift to the "operating procedures" themselves. As physical AI matures, the focus will move from hardware to the embedded skills and workflows—human expertise such as welding, surgery, and cooking can be encoded into scalable "robot operating manuals," forming a new form of intellectual property. Unitree's UniStore and NEURA Robotics' NeuraGym/NeuraVerse are considered early examples of the commercialization of such skills, although they currently rely on humanoid robot hardware manufacturers' platforms.

Barclays believes that in the extreme case of this trend, hardware will gradually become commoditized, and the real competitive advantage will belong to the party that can capture, encode, and control high-value skills—the company that builds the most robots may not be the ultimate winner, but the company that controls the "command layer" may dominate this new skills war.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.