The early predictions about AI and jobs — published roughly between 2022 and 2024 — focused heavily on creative and knowledge work. Three years into real enterprise adoption, the actual displacement pattern looks quite different from those projections. Some categories are seeing meaningful headcount pressure; others are seeing productivity gains that have, paradoxically, increased hiring.
The clearest displacement has come in specific narrow categories: first-tier customer support, basic copywriting and content production, entry-level paralegal document review, and certain categories of software development testing and QA. These were occupations where AI tools rapidly reached or exceeded human performance on the bulk of the work, and where the marginal cost of additional AI capacity is near zero. Hiring in these categories is down 20% to 45% from 2023 levels.
Software engineering, frequently predicted to be the most disrupted field, has had a more nuanced experience. Senior engineering productivity has increased substantially — most studies place the gain between 25% and 55% depending on the task mix — but enterprise demand for software has expanded in parallel. Net engineering hiring is roughly flat to slightly up, but the composition has shifted significantly: senior engineers are in higher demand than ever, while entry-level engineering hiring has dropped by roughly 30% as employers consolidate work in smaller, more senior teams.
Marketing, design, and content roles have followed a similar pattern. Headcount is down at the production end of these functions — junior designers, junior copywriters, basic asset producers — but is up at the strategic and senior end. Roles that combine creative judgment with AI tool fluency are paying meaningful premiums over equivalent roles three years ago.
Healthcare, perhaps surprisingly, has seen almost no displacement and substantial productivity gains. Radiology AI has not displaced radiologists; it has expanded the volume of scans they can read and improved diagnostic accuracy. Clinical documentation AI has reduced administrative burden, which has reduced burnout and turnover. Demand for clinicians continues to outpace supply by a wide margin.
Skilled trades remain entirely insulated. There is no near-term AI threat to electrical work, plumbing, HVAC installation, or any other category requiring physical presence and judgment in unstructured environments. If anything, the rising productivity of office workers has increased demand for skilled trades that serve the residential market, as higher-earning workers invest more in home improvement and infrastructure.
The roles most likely to be created by AI adoption, rather than displaced, are concentrated in three categories: AI deployment and engineering roles within enterprises (rapid growth), data infrastructure roles that support AI workloads (rapid growth), and AI safety, governance, and audit roles (small but growing rapidly). The total job creation in these categories is meaningful but is unlikely to fully offset displacement at the entry-level production end of the affected fields.
For workers, the practical guidance is consistent: invest deliberately in becoming the person who directs AI tools rather than the person whose tasks they replicate. The senior end of nearly every affected field is more in-demand and better-paid than ever. The entry-level end is genuinely harder, which is shifting the career-development question from 'how do I learn AI' to 'how do I get senior fast enough to be on the right side of this transition.'