AI Is Changing Job Demand Before Organizations Redesign Work

The decision is not whether to offer more AI training. It is whether the work, oversight and development system still gives people a credible way to build judgment as GenAI changes what one person can produce. Many organizations risk treating a work-design problem as an employee-skill problem. The business leader usually owns that review with the CHRO or CPO. The CLO should be involved when the review finds a capability gap rather than an institutional constraint.

The Texas signal makes that question harder to defer. Two-thirds of firms surveyed now use AI in their business processes, up from 40 percent two years earlier. Over the same period, job postings fell for occupations with tasks more exposed to GenAI: 5 percent by the end of 2023 and approximately 8 percent by the first quarter of 2025. Recent college graduates have experienced worse labor market outcomes, and current students have changed their educational decisions.

Those figures do not show that AI is reducing employment overall. They do show that demand is changing first around work whose tasks can be automated. I would look beyond the job title and ask what the person is still expected to learn by doing the work. If routine research, drafting or analysis becomes easier to automate, removing those tasks may also remove the practice through which employees develop the expertise needed for harder decisions. An organization can lower near-term demand while weakening its future supply of people who can exercise independent judgment.

People can use AI faster than organizations can update the work around them. The job description and approval route may stay fixed while staffing and progression remain designed for the old work. Faster output does not create value if it remains trapped in a review process built for slower production. A role that still carries decision accountability also needs enough independent expertise to question an AI-generated recommendation. Some controls, including expert approval and separation of duties, may be protecting the organization and should remain. The question is whether those controls still fit the changed work.

I would start with the work, not a course catalogue. For each role affected by adoption, identify the task that has changed, the decision that now needs human judgment, the evidence a person must be able to test independently, and the supervised practice that will build that capability. Then check whether staffing, approval and progression decisions support that route. If the constraint is an unchanged process or unclear decision right, training is not the first fix. If the revised role genuinely requires knowledge or judgment employees do not yet have, targeted learning and supervised practice may be part of the response.

The boundary is important. The Texas figures are an early regional signal, not proof of a uniform employment pattern or a general loss of jobs. AI-exposed work may remain in demand when adoption complements employees, increases the value of adjacent skills or creates more need for people who can frame problems, check outputs and apply domain judgment. Adoption rates alone cannot show whether firms are removing work, augmenting it or redesigning the surrounding process.

The judgment should change if longer-term evidence shows organizations consistently redesigning jobs, hiring, development routes and employment outcomes as adoption increases across more regions and occupations. Until then, the leadership test is more specific: can the revised work produce value while preserving a real path for people to acquire and exercise the judgment the organization still holds them accountable for?