Finishing a report faster does not settle what happens to the profession behind it. It changes the economics of a task. If that task also used to train a new colleague, the change reaches further into how a team develops experience. This is why AI’s impact on work needs more than a single jobs-gained or jobs-lost headline.
The pressure can arrive through fewer openings
Stanford Digital Economy Lab’s August 2026 update uses payroll records covering millions of US workers through June. Employment among 22–25-year-olds in AI-exposed occupations was 19% below the level it would have reached had it kept pace with less-exposed peers. The authors describe reduced hiring, rather than increased separations, as the main channel. They do not find economy-wide displacement.
That distinction matters to someone starting a career. A company can retain experienced people while opening fewer junior positions. A team can also raise output while reducing the routine assignments through which newcomers once learned the work.
A rebound may favour experience
Indeed Hiring Lab reports that senior roles accounted for 71% of the net increase in US software-development postings between May 2025 and May 2026. This describes advertisements, not completed hires. The analysis also places the rebound against a level still below the pre-pandemic baseline.
SignalFire’s 2026 talent report finds new-graduate and entry-level hiring at the large technology companies it tracks roughly 65% below 2019. These sources use different populations and measures. Their percentages should not be combined into a universal unemployment rate. Read together, they make a better case for looking at career stage separately from overall sector demand.
Make preparation observable
A practical starting point is to break a job into activities: gathering information, drafting, checking, agreeing requirements and making the final decision. Choose one small task without sensitive data for an AI-assisted trial. Measure the time spent on corrections and review as well as the initial output. Keep the quality requirement fixed so that a faster draft is not mistaken for a faster finished job.
A portfolio can then explain the decision process, not merely display the result. What need did the work address? Which option was rejected, and why? How was the output checked? Those details make a person’s contribution visible even when a tool produces much of the first draft.
Productivity planning also needs a learning route
For a manager, automating routine work creates a second design task: deciding how a newcomer will learn its underlying logic. Comparing examples, classifying errors or checking an output against a user requirement are possible assignments. These are proposed organisational responses, rather than outcomes measured in the cited studies.
Grant’s assessment
Who needs this, and for what? That question is more useful than assuming every profession will follow one trajectory. The evidence points to differences by experience, so preparation should combine tool fluency with problem selection and review. The next test of that view is whether actual hiring across career stages broadens, not simply whether more advertisements mention AI.
