Workplace

The WEF's numbers describe the job market in aggregate, but a 2025 Stanford study using detailed U.S. payroll data found something more specific: employment for 22-to-25-year-olds in the most AI-exposed occupations — like software development and customer service — has fallen nearly 20% below where it would be if it had kept pace with hiring for the same jobs among less AI-exposed peers, while employment among more experienced workers in the same fields shows no comparable gap [2]. The drop appears to come almost entirely from reduced hiring, not from firing existing staff — companies aren't laying off juniors, they're simply hiring fewer of them [2]. The researchers' leading explanation is that AI has gotten very good at the kind of knowledge you can learn from books and training data, but not at the tacit, on-the-job knowledge someone normally picks up by working alongside a senior colleague for a few years — which raises a harder question: if fewer people get that early apprenticeship, who trains the next generation of senior employees at all?

One of the most carefully measured workplace AI studies followed over 5,000 customer support agents using a generative AI assistant: overall productivity rose 13.8%, mostly from resolving issues faster [2]. The gain wasn't even — newer agents saw productivity jump 35%, reaching in two months what previously took six, while the most experienced agents saw little change [2]. Researchers read this as the AI spreading top performers' techniques to everyone else, like a fast, informal training program. That's a genuine win for narrowing skill gaps, though it raises a real question about what happens to the value of hard-won expertise when a tool can hand a novice most of it on day one. Worth holding next to the entry-level data above: AI can accelerate how fast a newly hired junior employee learns, but it may also be part of why fewer junior employees get hired to begin with. The two effects pull in opposite directions, and it isn't yet clear which one wins out over time.

When researchers tested three large language models screening over 500 resumes, white-associated names were preferred in 85.1% of comparisons showing any preference, versus 8.6% for Black-associated names [4]; separately, a University of Washington study found human reviewers tend to adopt whatever bias an AI hiring tool shows them, following even severely biased recommendations about 90% of the time [5]. At the same time, a different dynamic is reshaping hiring on both sides of the table: an estimated 57% of employers say candidates are using AI noticeably more in applications, and 90% report a spike in low-effort, AI-generated applications flooding their pipelines [6]. Employers have responded with AI of their own — resume filters, algorithmic scoring, AI-conducted interviews, which 63% of applicants say they've now experienced [7]. The result looks like an arms race that serves neither side well: recruiters now review roughly 291 applications per hire, nearly three times the 2021 rate, and it takes about 25% longer to fill a position than before the pandemic [7], while 38% of candidates have withdrawn from a role specifically because it involved an AI interview [7]. One argument worth weighing: if AI-polished writing makes every application sound similar, direct human relationships and networking may end up mattering more, not less, than they did before AI entered hiring.

New York City's Local Law 144 requires employers using an automated tool in hiring or promotion decisions to have it independently bias-audited every year, publish a summary of results, and give candidates 10 business days' notice before using it [6]. It's one of the first laws of its kind in the U.S. and is being watched closely — as a model or a cautionary tale, depending on who's asked — by other cities and states considering similar rules.

The 2023 Writers Guild of America and SAG-AFTRA agreements, reached after industry-wide strikes, built in some of the most specific AI protections in any U.S. labor contract to date: written consent and compensation before a studio can train AI on a writer's work or use a performer's digital likeness [7][8]. It's a model shaped by one industry's particular leverage and may not transfer directly elsewhere, but it shows that AI's role in a workplace is something workers can negotiate, not something handed down as fixed.

Some employers are experimenting with "internal talent marketplaces" — matching people to projects based on specific skills rather than a single fixed job title, so employees move between projects as needs shift rather than staying locked into one role [11]. Advocates argue this could ease the entry-level pipeline problem by making it easier to bring junior people in for discrete pieces of work. Microsoft's 2025 Work Trend Index describes a shift toward "human-agent teams" that assemble project-to-project rather than staying in fixed roles, comparing it to how a movie production hires a fresh team for each film [12]. The report illustrates this with one company — an AI-first advertising agency whose co-founder says they no longer need a strategist on every project because staff can access that expertise through an internal AI platform. It also cites a Harvard Business School field study of nearly 800 P&G employees, which found that a person working with AI can outperform a team without one, and that AI-assisted teams did a better job blending technical and business perspectives across departments [13].

Give groups the resume-screening statistics without naming the study. Have them predict whether the numbers seem realistic or exaggerated, then reveal the methodology (three LLMs, nine occupations, 554 resumes, ~40,000 comparisons). Discuss what made it convincing or not, and how they'd design a similar test.

Split the class: one group researches Local Law 144, the other the WGA/SAG-AFTRA AI provisions. Each presents how their mechanism works, whom it protects, and its biggest limitation. Close with a debate: for an industry of the students' choosing, would a government mandate or a negotiated labor agreement do more?

Give students the claim as Microsoft's Work Trend Index presents it — "companies are shifting toward hiring people for skills and assembling them into project-based teams, the way a movie production assembles a crew for one film" — without revealing the sourcing yet. In small groups, have them read the actual "human-agent teams" section of the report and list every piece of evidence it offers for that claim. Then have them sort what they find into two piles: evidence that directly supports the claim, and evidence that supports something narrower or different. They should land on the fact that the Supergood example is one company's self-description of how it operates, while the Harvard Business School field study of ~800 P&G employees is a real experiment — but about AI improving individual and team output, not about companies shifting to project-based hiring. Close with groups presenting their sort and a class discussion: should a report be allowed to use a rigorous study to lend credibility to a claim the study doesn't directly support? Where else have they seen this pattern — in the news, in advertising, in their own research?