Workers face new reskilling challenge as AI flattens skill demand across firms

Artificial intelligence (AI) is changing the structure of work across modern labor markets, prompting employers to rethink the skills they need as routine tasks become easier to automate and complex tasks become easier to augment. New research suggests that AI’s effect on employment may be less about simple job destruction and more about a deeper restructuring of skill demand inside firms, with China offering a large-scale example of how the shift is unfolding.

The study, titled Toward Sustainable Workforce Development: How AI Reshapes Skill Demand Structure: Evidence from 67 Million Job Postings in China, was published in Sustainability. Based on roughly 67 million online job postings from China between 2019 and 2024, the research provides evidence that AI exposure is associated with broader but shallower skill portfolios.

AI is changing the shape of skill demand, not just the number of jobs

AI’s labor market impact cannot be measured only by whether jobs disappear or survive. It is altering the content of work. Firms are changing the balance of skills they seek, the depth at which those skills are required and the way different skill categories combine within job roles.

The study separates AI exposure into two forces. The first is displacement, where AI overlaps with routine and codifiable tasks. The second is augmentation, where AI supports nonroutine tasks that involve judgment, analysis or complex decision-making. This distinction is important because the two forces do not move skill demand in the same direction.

• Displacement exposure is linked to weaker demand for routine cognitive skills, including rule-based information processing, writing, monitoring, mathematics and structured learning tasks. These are areas where AI tools can increasingly assist with drafting, classification, calculation, review and other repeatable knowledge-work processes.
• Augmentation exposure, on the other hand, is linked to higher demand for nonroutine analytical skills. These include critical thinking, complex problem solving, judgment, systems analysis, programming, operations analysis and troubleshooting. In these areas, AI does not simply replace workers. It changes the nature of the job by making workers more dependent on interpretation, oversight, problem framing and decision-making.

The study uses China as an example because of the scale of its online recruitment market and the speed with which AI-related technologies are diffusing across firms. However, the underlying pattern has wider relevance. In any economy where AI is entering workplaces, employers may increasingly reduce demand for narrow routine expertise while raising demand for workers who can coordinate, evaluate and apply AI-supported outputs.

The research also shows that AI’s impact is not limited to high-skill or white-collar work. Routine manual demand rises under both displacement and augmentation exposure in the Chinese data, suggesting that many operational, equipment-based and physical-presence tasks remain necessary even as digital systems reshape office and analytical work. This finding complicates the common assumption that AI will only shift labor demand upward toward more advanced roles.

Broader skill portfolios may come with lower depth

The study primarily focuses on skill depth. Many labor market analyses measure whether demand rises or falls for a skill category. This research asks whether employers demand those skills at higher or lower levels of importance.

Both displacement and augmentation exposure are associated with lower average skill importance. That means AI-exposed firms may be asking for a wider set of skills, but not necessarily demanding deep mastery in each one. The result is a flatter skill structure, where employees are expected to work across more areas but may have fewer opportunities to build a strong professional core.

The authors describe this as "de-coring." Practically, the term means that work may become broader, more flexible and more cross-functional, while also becoming less anchored in a single high-value expertise area. That creates a new challenge for workers. Career security has often depended on developing depth in a field, building firm-specific knowledge and gaining bargaining power through specialized capability. If AI weakens the value of deep single-track specialization, workers may need to constantly maintain a wider range of portable competencies.

This does not mean expertise is no longer needed. Rather, the study suggests that expertise is being reorganized. A worker may need enough technical, analytical, communication and operational ability to work with AI systems, but the role may no longer require the same depth of manual execution or narrow technical command that it once did.

Training systems that prepare workers for narrowly defined occupations may fall behind if employers increasingly demand adaptable skill mixes. At the same time, basic digital literacy will not be enough. Workers need broader competence, but also enough judgment and domain understanding to use AI effectively.

The findings also warn against reading skill demand too simply. A skill category can grow in share while declining in depth. For example, firms may post more jobs requiring analytical skills, but expect those skills at a lower level because AI tools handle part of the technical burden. Similarly, a firm may post fewer routine cognitive roles, but demand higher expertise from the remaining workers in that category.

This matters for policymakers and employers because headline job counts can hide major changes in job quality and career structure. A labor market may appear stable in terms of employment while workers experience deeper uncertainty in how their skills are valued, how they progress and how they remain employable.

China offers a warning for global workforce policy

China’s labor market provides a large-scale example of how AI-driven skill restructuring can unfold in an economy with fast digital adoption, large online recruitment channels and strong pressure for workforce upgrading. Although the study’s evidence comes from China, its lessons apply more broadly to economies trying to prepare workers for AI-era employment.

The research finds that de-coring is concentrated among small firms and firms with lower entry thresholds. These firms tend to require lower levels of education and experience and offer lower wages. That makes them especially important from a policy perspective. Workers in these firms may have the least access to employer-funded training, the weakest bargaining position and the fewest resources to adapt to rapid change.

This pattern suggests that AI may widen gaps between workers and firms. Large companies may have more capacity to invest in training, redesign jobs and use AI to improve productivity. Smaller firms may rely on AI to reduce costs and simplify roles, which could leave workers with broader task burdens but weaker career ladders.

In addition, displacement exposure is linked to lower education requirements, lower experience requirements and lower median salary in posted jobs. That suggests AI may push some firms toward lower-threshold recruitment, especially where technology can absorb parts of the work that previously required deeper training. Augmentation exposure shows a somewhat different pattern, with weaker links to education requirements but negative associations with experience and salary.

For governments, the study suggests that workforce development cannot focus only on creating more AI specialists. It must also support the much larger group of workers whose roles are being changed by AI without becoming AI jobs. These workers need training that is modular, portable and practical across occupations.

The study points toward several policy priorities:

• Training systems should support broad, transferable competencies rather than locking workers into narrow occupational tracks.
• Credentials should be portable across firms and sectors.
• Small firms may need public support through training vouchers, sectoral skill programs or shared workforce development platforms.
• Workers may also need long-term learning accounts that allow them to build skills across multiple employers over time.

The research also raises questions for employers. Firms can use AI in ways that hollow out jobs, reduce skill depth and weaken career progression. However, they can also use AI to redesign work around human judgment, supervision, creativity and problem-solving. The difference depends on whether companies treat AI as a cost-cutting substitute or as a tool for building more capable teams.

Speaking of the study limitations, it measures potential AI exposure based on the overlap between AI patent texts and occupational task descriptions. It does not prove that every firm in the sample adopted AI tools. It also focuses on online job postings, which are more likely to capture formal, urban and internet-mediated recruitment than informal or rural work.