Career Capital in the AI Era: An Evidence-Based Guide for Professionals

Career advice about artificial intelligence has an evidence problem. Most of what circulates is either a confident prophecy about mass displacement or an equally confident reassurance that nothing important will change. Both are cheap to produce and difficult to act on.

The research picture is more interesting, and more useful. Since late 2024 a body of large-scale empirical work has emerged that measures what is actually happening in labour markets rather than what might happen. Taken together it suggests something counter-intuitive: aggregate employment and earnings have barely moved, while the internal structure of professional careers — how people enter occupations, how they accumulate expertise, and what employers pay a premium for — is changing measurably. The disruption is showing up in composition long before it shows up in headline statistics.

That distinction matters if you are making decisions about your own career. It means the useful question is not "will AI take my job" but "which parts of my work still build career capital, and which have quietly stopped."

This guide is written for working professionals — individual contributors, managers, specialists and those advising them — across knowledge-intensive sectors. It examines what the evidence supports, what remains contested, and what is still speculation. It then offers a practical framework for auditing your own position and a set of concrete actions calibrated to career stage.

This article is a synthesis of published research and primary sources, cited throughout. It does not draw on proprietary data or first-hand consulting engagements, and it flags the strength of the evidence behind each claim.

---

Executive summary

• Aggregate effects remain small; compositional effects do not. Rigorous administrative-data studies in Denmark find precisely estimated null effects on earnings and hours roughly two years after ChatGPT's release (Humlum & Vestergaard, 2026), while US payroll data show sharp declines concentrated in specific groups (Brynjolfsson, Chandar & Chen, 2025).
• Early-career workers in AI-exposed occupations are the clearest pressure point. Workers aged 22–25 in the most AI-exposed occupations showed a 16% relative employment decline in US payroll records, with adjustment occurring through headcount rather than pay.
• Declines concentrate where AI automates rather than augments. Occupations where usage patterns skew toward full task delegation show employment declines; augmentation-heavy occupations show muted or positive changes.
• The apprenticeship channel is narrowing. Analysis of over one billion job advertisements found AI-exposed entry-level roles are seven times more likely to demand traditionally senior skills such as judgement and leadership (PwC, 2026).
• AI skills carry a large and growing wage premium — an average 62% in PwC's 2026 analysis, up from 57% the prior year — though premiums of this size are typical of early-stage scarcity and should not be assumed durable.
• Skills-based hiring is real in policy and thin in practice. Fewer than 1 in 700 US hires reflected the removal of degree requirements, with nearly all measurable change concentrated in about 37% of firms (Fuller, Sigelman & colleagues, 2024).
• AI compresses the novice–expert gap in output but not in judgement. A field study of 5,172 customer-support agents found a 15% average productivity gain concentrated among less-experienced workers — the same mechanism that erodes the learning value of junior tasks.
• Verification is becoming the core professional skill, and it is fragile: knowledge workers with higher confidence in AI report less critical engagement with its output (Lee et al., 2025).
• Regulation of AI in hiring is real but unsettled. The EU deferred its high-risk obligations for employment AI to 2 December 2027, and Colorado replaced its landmark AI Act in May 2026.
• The durable strategy is to convert AI-assisted throughput into demonstrable judgement — accountable decisions, verified outputs and portable evidence — rather than to accumulate tool familiarity alone.

---

What the evidence actually shows

The aggregate picture is quiet

The most methodologically careful national study to date comes from Denmark, where researchers linked two large representative worker surveys to administrative employer–employee records. Covering roughly 25,000 workers across 7,000 workplaces in eleven occupations widely considered exposed to generative AI, Humlum and Vestergaard estimate null effects on earnings and recorded hours, with confidence intervals tight enough to rule out effects larger than about 2% two years after ChatGPT's launch. Users themselves reported average time savings of only about 2.8% of work hours — far below the 15%–50% gains observed in controlled trials.

The authors' interpretation is important and often missed in coverage: the absence of an earnings effect is not evidence that nothing is happening. They argue that firms absorb the technology through task reorganisation — new work in content generation, AI oversight and integration — well before any of it reaches wage or hours data. Their revised paper title, referencing still waters and rapid currents, captures the point.

Evidence strength: high for the null result in the Danish setting; moderate for generalisation, since Denmark's labour market is flexible but its institutions, wage bargaining and firm mix are not identical to those of the US, UK or emerging economies.

The composition is not

The countervailing finding comes from US payroll administrative data. Brynjolfsson, Chandar and Chen analysed records from the largest US payroll provider and found that since generative AI's widespread adoption, workers aged 22–25 in the most AI-exposed occupations experienced a 16% relative employment decline, even after controlling for firm-level shocks. Employment for more experienced workers in the same occupations, and for workers in less exposed fields, remained stable or grew.

Three details make this more credible than a simple correlation:

1. Adjustment ran through employment, not compensation. Firms reduced junior headcount rather than cutting pay — consistent with hiring-margin adjustment rather than a general demand shock.
2. Declines concentrated in automation, not augmentation. Where usage data indicated full task delegation, entry-level employment fell; where usage indicated human–AI collaboration, changes were muted.
3. The authors tested competing explanations. A February 2026 follow-up specifically addressed whether interest-rate movements explained the pattern better than AI exposure, and concluded they did not.

Evidence strength: moderate to high for the descriptive fact; moderate for causal attribution. The sample is large but not nationally representative, the exposure measures are proxies, and the paper is a working paper rather than a peer-reviewed publication. It should be read as a strong early signal, not a settled result.

The wage signal

PwC's 2026 Global AI Jobs Barometer, released 15 June 2026, analysed more than one billion job advertisements across 27 countries. It reports an average wage premium of 62% for jobs requiring specific AI skills, up from 57% the prior year, and finds AI-skill postings growing at 69% against 9% for the overall jobs market.

Read this with care. Large premiums are characteristic of scarce, newly defined skill categories, and they typically compress as supply catches up — the same pattern seen with earlier waves of specialist software and data skills. The premium is a real market signal about current scarcity; it is not a forecast of durable returns.

Evidence strength: moderate. Job-advertisement data is large-scale and timely but measures stated employer demand, not realised hiring or pay. It is also commercial research rather than peer-reviewed work.

---

Why the career ladder is changing shape

The most consequential finding for individual professionals is not about job losses. It is about how expertise gets built.

For decades, professional careers ran on an implicit exchange: junior workers performed high-volume, low-judgement tasks — document review, first-draft memos, reconciliation, basic code, research summaries — and in return absorbed pattern recognition, domain context and tacit standards. The work was economically marginal but developmentally essential.

Generative AI is unusually good at exactly that band of work. The field evidence is direct: in a study of 5,172 customer-support agents published in the Quarterly Journal of Economics, access to an AI assistant raised issues resolved per hour by 15% on average, with less-experienced and lower-skilled workers improving in both speed and quality while the most experienced saw small speed gains and small quality declines. Treated agents with two months of tenure performed comparably to untreated agents with more than six months.

That is a genuinely positive finding about individual capability. It is also, from an employer's perspective, an argument for buying fewer months of tenure — and from a career perspective, an argument that the rungs are being removed.

The job-advertisement evidence is consistent. PwC's analysis of 2.4 million US entry-level postings found that AI-exposed entry-level roles are seven times more likely to require traditionally senior "human-intensive" skills such as leadership, creativity and face-to-face interaction. Openings for these "seniorised" entry-level roles grew 35% since 2019, while other entry-level roles shrank 10%. PwC's global workforce leader described AI as "removing some of the routine work that once acted as an apprenticeship."

Employer surveys point the same direction, with the usual caveats about self-report: ZipRecruiter's 2026 AI Employer Report, based on a survey of more than 1,000 US talent-acquisition professionals conducted 11–18 June 2026, found 31% saying AI has raised experience requirements for entry-level roles.

Table 1 — How career-building mechanisms are shifting

Original table developed for OneWise from the sources cited in this article.

Callout — a tension worth holding The same mechanism that makes AI valuable to you as an individual — it lifts your output toward expert level quickly — is what makes junior roles harder to justify economically. Personal benefit and structural risk come from the same source. Career strategy in this period is largely about managing that tension deliberately rather than pretending it away.

---

A framework for career capital under AI

The following model was developed for this article. It is an organising device grounded in the evidence above, not a validated instrument.

The three-layer career capital stack

Layer 1 — Executional capacity. What you can produce: drafts, analyses, code, designs, models. This is the layer AI most directly substitutes for and where the novice–expert gap is compressing fastest. It remains necessary. It is increasingly insufficient as a differentiator.

Layer 2 — Accountable judgement. What you can decide and defend: which analysis is wrong and why, what risk is acceptable, which trade-off to take, when the confident output is subtly incorrect. This is where the evidence points as the appreciating asset — both in PwC's "professionalised" role category and in Autor's argument that AI's opportunity is to extend the reach and value of human expertise. Autor is careful about epistemic status, writing that his thesis "is not a forecast but an argument about what is possible."

Layer 3 — Situated context and trust. What only you know about this organisation, market, regulator, client or codebase, plus the relationships that let you act on it. This is the least substitutable layer and the least portable — which makes it valuable and simultaneously a lock-in risk.

Most stalled careers in this period will be over-invested in Layer 1 and under-invested in Layer 2.

Figure 1 — The Career Capital Stack Purpose: Show the three layers, their relative substitutability by AI, and where professionals should shift investment. Components: Three horizontal bands stacked vertically. Bottom band, widest: "Executional capacity — producing output." Middle band: "Accountable judgement — deciding and defending." Top band, narrowest: "Situated context and trust — knowing this specific environment." Labels and hierarchy: A vertical gradient arrow on the left, labelled "AI substitutability," running dark at the bottom (high) to light at the top (low). A second arrow on the right, labelled "Career differentiation," running the opposite direction. Each band carries two or three example activities in smaller type. Caption: "AI substitutes most readily for the layer that historically absorbed the most professional time. Differentiation is moving upward."

The task portfolio audit

Before changing anything, map your actual week. List your recurring tasks, then place each one on two axes: how much of your time it consumes, and how substitutable it is by current AI tools in your specific context — including data access, confidentiality constraints and quality tolerance.

Table 2 — Task portfolio quadrants and the appropriate response

Original framework developed for OneWise.

Figure 2 — Task Portfolio Matrix Purpose: Give readers a reusable two-by-two for auditing their own work. Components: X-axis, "AI substitutability in your context" (low → high). Y-axis, "Share of your working time" (low → high). Four quadrants labelled Defend (low/low), Deepen (high/low), Supervise (low/high), Delegate (high/high). Labels: Each quadrant contains a one-line instruction. Three or four small plotted dots as illustrative example tasks, with a note clarifying they are illustrative only. Caption: "Substitutability is context-specific. The same task can sit in different quadrants at two organisations."

---

The skills that are appreciating — and how to read the claims

Skills forecasts deserve scepticism, because they are usually employer expectations rather than measured outcomes. The World Economic Forum's Future of Jobs Report 2025, published 8 January 2025 and drawing on more than 1,000 employers representing over 14 million workers across 55 economies, projects that 39% of workers' core skills will change by 2030 — down from 44% in its 2023 edition. It also reports that 63% of employers cite skills gaps as the primary barrier to business transformation.

These are forecasts, and the downward revision from 44% to 39% is itself a reminder that such projections move. They are useful as a read on employer intent, not as a prediction of what will happen.

Table 3 — Claim, evidence type and confidence

Original table developed for OneWise.

The category with the most consistent support across independent sources is verification and judgement under uncertainty. A CHI 2025 study by Lee and colleagues at Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers who supplied 936 real examples of AI use at work. It found that higher confidence in the AI tool was associated with less critical thinking, while higher confidence in one's own ability was associated with more. Qualitatively, the nature of the work shifted toward verification, integration and stewardship.

That is a self-reported, correlational finding and should not be over-read. But it identifies a plausible failure mode with a clear professional implication: the ability to catch a confident, fluent, wrong answer is becoming a core competence, and it degrades precisely when the tool feels most reliable.

---

Credentials, hiring and the verification gap

Professionals repositioning themselves often assume employers will assess them on demonstrated skill. The evidence urges caution.

A joint study by Harvard Business School's Project on Managing the Future of Work and the Burning Glass Institute, published 14 February 2024, examined 11,300 roles at large US firms before and after degree requirements were removed. About 3.6% of roles dropped a requirement. Within those roles, hiring of workers without a bachelor's degree rose about 3.5 percentage points — but netted against the small share of affected roles, the overall shift was roughly 0.14 percentage points, or fewer than 1 in 700 hires. Nearly all of the measurable change occurred within about 37% of firms studied.

The practical reading for a professional is not that credentials are permanent. It is that stated policy is a weak predictor of screening behaviour, and that the burden of making capability legible falls on the candidate. Portfolio evidence, verifiable project outcomes and referenceable work tend to travel better than assertions of AI fluency.

Figure 3 — The intent–practice gap in skills-based hiring Purpose: Illustrate why announced policy changes translate into very small hiring shifts. Components: A left-to-right funnel with four stages: "Policy announced" (widest), "Requirement removed from posting," "Screening process changed," "Non-degreed candidate hired" (narrowest). Beneath the funnel, a branch showing two paths from stage two: a thick arrow to "Screening unchanged → no measurable effect" and a thin arrow to "Screening redesigned → measurable effect." Labels: Annotate stage two with "≈3.6% of roles studied" and the endpoint with "<1 in 700 hires overall." Caption: "Removing a requirement from a job posting is not the same as changing how candidates are evaluated."

---

Regulation, and why it affects your job search

AI is now used at multiple points in hiring and workforce management, and the rules governing it are genuinely unsettled — which matters both for candidates and for anyone whose role touches people decisions.

• European Union. The EU AI Act (Regulation (EU) 2024/1689) classifies AI used in employment and worker management as high-risk under Annex III. Those obligations were originally due to apply from 2 August 2026. The Digital Omnibus on AI, which entered into force in late July 2026, deferred stand-alone Annex III high-risk obligations to 2 December 2027, and to 2 August 2028 for AI embedded in regulated products. Most Article 50 transparency obligations were not deferred and applied from 2 August 2026.
• Colorado, US. Senate Bill 26-189, signed 14 May 2026 and effective 1 January 2027, repealed and replaced the 2024 Colorado AI Act before it took effect. The replacement narrows the framework toward notice, adverse-outcome explanation and meaningful human review of automated decision-making technology in consequential decisions, including employment.
• Other US jurisdictions. California's Civil Rights Council regulations on automated-decision systems in employment took effect 1 October 2025, and Illinois and New York City maintain their own requirements. Obligations differ meaningfully by jurisdiction.

What this means practically: in some jurisdictions you may have a right to notice that an automated system influenced a decision about you, and in some, a route to human review. Do not assume those rights exist where you are, and do not assume the timelines above are final — this area has been repeatedly amended.

---

Latest developments

• 27 July 2026 — The EU's Digital Omnibus on AI entered into force, deferring high-risk obligations for Annex III systems, including employment and worker-management AI, from 2 August 2026 to 2 December 2027. Established fact.
• 27 June 2026 — Reporting on the expanded Stanford Digital Economy Lab–ADP Research partnership described the Canaries Dashboard as drawing on data covering approximately 4.6 million workers across more than 730 occupations, providing continuously updated indicators of AI-related labour-market change. Established, though the underlying sample is not nationally representative.
• 15 June 2026 — PwC published its 2026 Global AI Jobs Barometer, reporting the 62% AI-skills wage premium and the "professionalised versus democratised" two-track finding. Company research; directionally consistent with other sources but not independently replicated.
• 14 May 2026 — Colorado enacted SB 26-189, replacing its landmark AI Act. Established fact.
• March–April 2026 — Platform data from Handshake indicated 4.2% of full-time early-career postings mentioned AI skills as of March 2026, roughly double a year earlier, against broadly flat overall early-career posting volumes. Single-platform data; indicative, not representative.
• 9 February 2026 — The Canaries authors published a note testing whether interest-rate movements better explain the observed entry-level declines, concluding they do not. Preliminary research; addresses a live methodological dispute.

---

Frequently misunderstood ideas and common myths

"The data shows AI is destroying jobs." It does not, at least not yet in aggregate. The best-identified national study finds null effects on earnings and hours. What the data shows is redistribution — concentrated, real, and currently falling hardest on entry points.

"Learning to prompt is the skill." Prompting is a tool interface, and interfaces change. The transferable capability is problem specification, verification and knowing when the output is wrong — which is domain knowledge, not tool knowledge.

"Employers have moved to skills-based hiring." Announcements have moved. Hiring outcomes have moved far less.

"Senior roles are safe." The declines are concentrated among junior workers so far. That is an observation about the present, not a structural guarantee. The same field evidence showing large novice gains also showed small quality declines among the most skilled — a hint that senior workflows are not automatically improved by AI either.

"If I don't adopt AI I'll be replaced by someone who does." Widely repeated, weakly evidenced at the individual level. The Danish data found no earnings effect for adopters. Adoption may still be prudent; the specific promise of individual earnings gains is not currently supported.

---

Mistakes to avoid

1. Optimising for throughput. Producing more with AI is easily matched by peers and rarely differentiating. Producing defensible work is harder to replicate.
2. Letting the delegated tasks be the ones you never learned. If AI is drafting your models or memos before you can build them unaided, you are trading present efficiency for a permanent gap in judgement.
3. Treating tool certificates as career capital. They signal exposure, not capability, and depreciate with each model release.
4. Assuming your Layer 3 context transfers. Deep organisational knowledge is valuable and largely non-portable. Balance it with externally legible work.
5. Over-indexing on a single study. Both the alarming payroll findings and the reassuring Danish nulls are early, partial and contested.
6. Waiting for certainty. The compositional shifts are measurable now; the aggregate resolution may take years.

---

Practical takeaways

Table 4 — Actions by career stage

A 90-day sequence

1. Weeks 1–2: Complete the task portfolio audit. Write it down; do not do it from memory.
2. Weeks 3–4: Identify which learning your Delegate quadrant used to generate, and where that learning will now come from.
3. Weeks 5–8: Build one piece of externally legible evidence of judgement — a decision memo, a documented review process, a public write-up, a reviewed contribution.
4. Weeks 9–12: Establish a verification routine you can describe to an interviewer, and test one adjacent capability from your Defend quadrant.

---

Key insights

1. Aggregate labour-market statistics currently understate the change; composition is moving faster than levels.
2. The clearest measured effect is on entry-level employment in AI-exposed occupations, not on incumbents.
3. Employment declines concentrate where AI automates tasks rather than augmenting workers.
4. AI's largest productivity gains accrue to novices — which is both an individual opportunity and a structural risk to apprenticeship.
5. Entry-level roles are increasingly asking for senior human skills, compressing the traditional ladder.
6. The AI-skills wage premium is large now; scarcity premiums historically compress.
7. Skills-based hiring remains far more prevalent as policy than as practice.
8. Verification of AI output is becoming a core professional skill, and confidence in the tool tends to reduce it.
9. Regulation of employment AI is real but volatile; timelines have already shifted more than once.
10. The most defensible career asset is accountable judgement that others can verify — not tool fluency, which depreciates.

---

Frequently asked questions

Is AI actually causing job losses right now? In aggregate, the best-identified evidence finds no measurable effect on earnings or hours. In specific segments — particularly workers aged 22–25 in the most AI-exposed occupations — US payroll data show substantial relative declines.

Which jobs are most exposed? Exposure tracks tasks, not job titles. Roles concentrated in routine information processing, drafting, summarising and standardised analysis show the highest exposure. Roles requiring physical presence, negotiation, accountability or deep situated context show the least.

Should I put "AI skills" on my CV? Only with specifics. "Used AI tools" signals little. "Redesigned a review workflow that cut turnaround by X while maintaining error rate" signals judgement. Job-advertisement data shows genuine demand for AI skills, but employers increasingly want evidence of applied outcomes.

Are AI certifications worth it? They can help with screening filters and structured learning, but they depreciate quickly and signal exposure rather than capability. Prioritise them below demonstrable work.

Is it still worth getting a degree? The evidence suggests degree requirements have changed far less in practice than in public commitments — fewer than 1 in 700 hires reflected the shift. That is an argument for realism about screening, not an endorsement of any particular educational path.

What should early-career professionals do differently? Protect learning that automation would otherwise skip. Do foundational work unaided at least once before delegating it, seek stakeholder-facing exposure earlier than previous cohorts did, and accumulate verifiable outcomes.

Does AI help or hurt experienced professionals? Field evidence found small speed gains and small quality declines for the most skilled workers on a narrow task. That is one study in one occupation, but it cautions against assuming seniority guarantees benefit.

Is the wage premium for AI skills going to last? Unknown. Large premiums for scarce new skills have historically compressed as supply expands. Treat the current premium as a signal about today's market rather than a projection.

How do I avoid becoming dependent on AI? Deliberately practise core tasks unaided at intervals, keep a verification routine, and be alert to the finding that trusting the tool more is associated with checking it less.

What is the difference between automation and augmentation, and why does it matter? Automation means fully delegating a task; augmentation means collaborating on it. Employment declines have concentrated in occupations where usage skews toward automation, making this distinction practically relevant when choosing roles.

Are employers allowed to use AI to screen my application? In many jurisdictions, yes, subject to conditions. The EU classifies employment AI as high-risk, with obligations now applying from 2 December 2027. Some US states require notice, explanation or human review. Rules vary substantially and have been repeatedly amended.

Can I find out whether AI was used to reject me? Sometimes. Several jurisdictions grant notice or explanation rights, but coverage, timing and enforcement differ. Check the specific rules applicable to the employer's location.

Should I switch industries to avoid AI exposure? Rarely a good reason on its own. Exposure varies more by task mix than by sector, and switching resets accumulated situated knowledge, which is among the least substitutable assets you hold.

What is a "seniorised" entry-level role? A junior position whose posting demands skills historically associated with senior roles — leadership, judgement, client-facing work. Analysis of US entry-level postings found these grew 35% since 2019 while other entry-level roles declined 10%.

How reliable is the research on this? Mixed and improving. The strongest work uses administrative data or randomised rollouts; much of the widely cited material is employer surveys or job-advertisement analysis, which measure intent rather than outcome. Several key papers remain unpublished working papers.

What single change matters most? Shift from being the person who produces the output to the person accountable for whether it is right — and make that accountability visible to people outside your organisation.

---

Glossary

• Augmentation — AI use in which a human and system collaborate on a task rather than the task being fully delegated.
• Automation — AI use in which a task is fully delegated to the system.
• Annex III (EU AI Act) — The schedule listing stand-alone high-risk AI use cases, including employment and worker management.
• Automated decision-making technology (ADMT) — Term used in recent US state law for systems that materially influence consequential decisions.
• Career capital — The accumulated, transferable assets that make a professional valuable: capability, judgement, context and relationships.
• Difference-in-differences — A statistical method comparing changes over time between exposed and unexposed groups to isolate an effect.
• Digital Omnibus on AI — The 2026 EU legislative package amending the AI Act, principally by deferring high-risk obligations.
• EU AI Act — Regulation (EU) 2024/1689, the EU's risk-based framework for AI systems.
• Exposure (occupational) — A task-based measure of how much of an occupation's work AI could plausibly perform; a proxy, not a prediction of displacement.
• Null effect (precisely estimated) — A finding of no measurable effect, with confidence intervals narrow enough to rule out effects above a stated size.
• Seniorised entry-level role — A junior posting requiring traditionally senior skills.
• Skills-based hiring — Evaluating candidates on demonstrated capability rather than credential proxies such as degrees.
• Situated knowledge — Context-specific understanding of a particular organisation, market or system; valuable and largely non-portable.
• Working paper — Research circulated before peer review; findings may change on revision.

References

Academic papers and peer-reviewed research

Autor, D. H. (2024). Applying AI to rebuild middle class jobs (NBER Working Paper No. 32140). National Bureau of Economic Research. https://doi.org/10.3386/w32140

Brynjolfsson, E., Chandar, B., & Chen, R. (2025). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence (Working paper, rev. 13 November 2025). Stanford Digital Economy Lab. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/

Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044

Humlum, A., & Vestergaard, E. (2026). Still waters, rapid currents: Early labor market transformation under generative AI (rev. of NBER Working Paper No. 33777). National Bureau of Economic Research. https://www.nber.org/papers/w33777

Conference papers

Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778

Industry and institutional reports

Brynjolfsson, E., Chandar, B., & Chen, R. (2026, February 9). Canaries, interest rates, and timing: More on the recent drivers of employment changes for young workers. Stanford Digital Economy Lab. https://digitaleconomy.stanford.edu/news/canaries-interest-rates-and-timinga-more-on-recent-drivers-of-employment-changes-for-young-workers/

Harvard Business School Project on Managing the Future of Work & The Burning Glass Institute. (2024, February 14). Skills-based hiring: The long road from pronouncements to practice. https://www.hbs.edu/managing-the-future-of-work/Documents/research/Skills-Based%20Hiring.pdf

PwC. (2026, June 15). 2026 Global AI Jobs Barometer. PricewaterhouseCoopers. https://www.pwc.com/gx/en/services/ai/ai-jobs-barometer.html

Stanford Digital Economy Lab. (2026). Canaries Dashboard. https://digitaleconomy.stanford.edu/project/indicators/canaries-dashboard/

World Economic Forum. (2025, January 8). The future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/

ZipRecruiter Economic Research. (2026). More jobs, higher bar: The 2026 AI employer report. https://www.ziprecruiter-research.org/economic-insights-research/ai-employer-report-2026

Legislation, standards and regulatory sources

European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj

Colorado General Assembly. (2026). Senate Bill 26-189: Automated decision-making technology. https://leg.colorado.gov/bills/sb26-189

Ongoing data resources

Federal Reserve Bank of New York. (n.d.). The labor market for recent college graduates. https://www.newyorkfed.org/research/college-labor-market

Anthropic. (2026, March 5). Labor market impacts of AI: A new measure and early evidence. https://www.anthropic.com/research/labor-market-impacts (Company-published research; cited for its measurement approach and flagged as non-independent.)

Note on sourcing: where this article reports figures from commercial research or employer surveys, that provenance is stated in the text. Working papers are identified as such. Reported findings reflect the versions available as of 4 August 2026 and may be revised.