AI workforce coverage across layoffs, hiring shifts, productivity studies, new job designs, entry-level work, and the skills practitioners need as AI changes teams.
Topic brief
Brief updated Aug 3, 2026
AI workforce covers how artificial intelligence is reshaping employment: which jobs are automated or augmented, how hiring and skills requirements change, how companies reorganize around AI agents, and how workers, unions, regulators and governments respond. It spans layoffs and restructuring, the talent market for AI expertise, pay differentials, reskilling and curriculum design, workplace surveillance rules, and the macroeconomics of productivity.
For practitioners and leaders, the questions are concrete: how to redesign roles and teams when agents can do parts of the work, how to hire and evaluate for AI fluency, how to measure real productivity gains against claims, and how to manage the human side of rapid change. The evidence remains mixed, with survey data showing role restructuring at the entry level while pay data shows the premium landing unevenly across technical roles. Measurement level matters more than most commentary admits: firm-level datasets, national labour statistics and hiring-manager surveys routinely disagree about the same phenomenon, so a claim about AI and jobs is only as good as the population it was drawn from.
The stakes are economic and political. Governments are debating safety nets and economic plans, companies compete for scarce AI talent while cutting other roles, litigation over AI-related discrimination is underway, and regulators are beginning to write rules on AI hiring and worker monitoring. How this plays out shapes wages, inequality and the social license for AI deployment.
The most consequential recent evidence points the AI-and-work question away from headcount and toward pay and task composition. Apollo's July 30 white paper by analysts Sania Edlich and Torsten Sløk estimated that real-wage growth in high-AI-exposure occupations ran 6.7 percentage points slower after 2023 than in lower-exposure occupations, with the estimated effect concentrated in service occupations and the bottom wage quartile and no statistically significant employment effect in the main model; exposure came from Anthropic's Economic Index, an occupation-level proxy built from observed use of one provider's tools rather than a record of employer deployment. Compensation datasets arrived at a similar place from a different direction. Payscale's 2026 Flight Risk Report found new Software Engineer V hires earning a median $230,000 against $194,000 for tenured peers, and Payscale's Ruth Thomas told Lets Data Science that AI fluency is already assumed in core tech and no longer moves pay, with 8 to 12% new-hire premiums surfacing instead in roles such as Marketing Analyst III and IT Product Manager IV where the capability is still scarce. Toptal's Q2 2026 report recorded a 28% year-over-year rise in experienced full-time data science postings against flat offered compensation, which Chief Economist Erik Stettler attributed to demand improving faster than candidate bargaining power in a market still holding a deep pool of laid-off applicants. Indeed Hiring Lab's July 14 finding of a 42% median hourly premium in data-center installation and maintenance postings, about $10 per hour, shows the same repricing reaching the trades, though it measures advertised pay across postings rather than realized earnings.
Two large usage studies published in the same window help explain why aggregate headcount is a lagging signal. Google's ATLAS study, released July 23 from about 15 million de-identified interactions, found AI activity across 68% of occupations representing 90% of U.S. employment but in only about 21% of tasks within a typical job, with fewer than 10% of workplace interactions fully automating a task. OpenAI's July 27 Work at the Frontier report found that 43.5% of occupation-specific messages in a sample of more than 800,000 concerned tasks historically associated with another occupation, while cautioning that the unit of analysis is a message and the data cannot show whether AI created new cross-occupation work or helped users perform responsibilities they already had. Where headcount did move, the stated rationale was organizational rather than substitutive: monday.com disclosed in a July 22 Form 6-K a roughly 20% workforce reduction affecting about 620 people and $45 million to $55 million in net restructuring charges, and said the reduction was not designed to replace people with AI but to reorganize around its AI Work Platform while continuing to hire in selected strategic areas. Readiness data suggests the operative constraint sits with managers: ManpowerGroup's July 22 study of 80 C-suite, CHRO and senior talent-acquisition leaders in the U.S. and U.K. found only 3% rated their leaders highly prepared to manage AI-enabled ways of working, and Indeed's July 28 survey of 1,301 U.S. respondents found 43% of managers felt unequipped to lead AI-native workers while 52% of workers said they were not receiving the AI training they needed. The measurement gap underneath all of these claims is itself now a legislative question: S. 4742, the AI DATA Act introduced June 10 by Senators Mark Kelly, Jim Banks and Todd Young, would add quarterly AI-adoption questions to a Census Bureau business survey for 10 years and require annual Labor and Commerce analyses, but it remains referred to the Senate Committee on Health, Education, Labor, and Pensions rather than enacted law.
S. 4742, the AI DATA Act, sits with the Senate Committee on Health, Education, Labor, and Pensions after its June 10 introduction; watch whether it advances, since it would add quarterly AI-adoption questions to a Census Bureau business survey for 10 years and require annual Labor and Commerce analyses that would make later displacement claims easier to test. Australia's Department of Employment and Workplace Relations found no broad AI disruption in its review of occupational change since November 2022, and ABC reports the government plans regular monitoring, so the follow-up reports and whether early weakness in exposed occupations such as telemarketers persists are checkable signals. In Victoria, Labor's July 20 pledge on AI hiring and workplace surveillance is an election commitment the party says it would legislate if returned in November, and the decisive event is publication of draft legislation showing definitions, exemptions and enforcement powers. The record retrieved for this audit does not establish whether the Wikimedia Foundation granted or denied the U.S. recognition request that set a July 24 response deadline, and the Meta discrimination case continues after a denial of temporary relief that decided nothing about the merits. monday.com said it expects its restructuring to be substantially completed in the second half of 2026 while continuing to hire in selected strategic areas, which is testable against later disclosures, as is Walmart's stated multi-year aim to give each of about 2.1 million associates some level of AI skill. Apollo's wage-growth estimate has not been replicated and the evidence does not establish that AI caused the gap it measured, so watch for replication against other model-use datasets, employer-level adoption measures and longer post-adoption periods, alongside follow-up work from London's AI and Jobs Taskforce and any concrete outputs from the Fed's Productivity and Jobs task force.
| source | stated limit | headline finding | what it measured |
|---|---|---|---|
| Apollo white paper (July 30, 2026) | Exposure is an occupation-level proxy based on observed use of one provider's tools, not a record of whether any employer deployed AI across a role | Real-wage growth 6.7 percentage points slower in high-AI-exposure occupations after 2023; no statistically significant employment effect in the main model | Difference-in-differences across 321 matched US occupations on a 2015-2025 panel, combining Bureau of Labor Statistics wage and employment data with Anthropic's Economic Index |
| OpenAI, Work at the Frontier (July 27, 2026) | OpenAI says the sample should not be generalized to ChatGPT Enterprise users or the entire U.S. workforce, and the study does not observe productivity, output quality or employment effects | 43.5% of occupation-specific messages concerned tasks historically associated with another occupation; 16.8% across all work-related messages | A random sample of more than 800,000 work-related messages from individual ChatGPT accounts of U.S. users, with occupations assigned from ChatGPT Business account information and task boundaries based on O*NET |
| Google ATLAS v1.0 (released July 23, 2026) | Axios reports Gemini Enterprise and Google Workspace activity was excluded because Google does not retain the required logs, so it covers selected Google products rather than the labor market | AI activity in 68% of occupations representing 90% of U.S. employment, but only about 21% of tasks in a typical job; fewer than 10% of workplace interactions fully automated a task | About 15 million aggregated, de-identified interactions across the Gemini app, Google Search's AI Mode and the Gemini API, mapped to occupational tasks |
| Australia's Department of Employment and Workplace Relations (reported by ABC) | ABC reports the government plans regular monitoring, and Minister Amanda Rishworth said AI could still reshape the jobs market, so this is a baseline rather than a settled result | AI is not yet causing broad disruption; labour-market conditions remain strong by historical standards, with early weakness in some occupations predicted to be highly exposed, including telemarketers | Occupational change in Australia's labour market since the release of ChatGPT in November 2022 |
| OECD analysis of London jobs (reported July 7, 2026) | The London.gov report emphasises that high exposure does not automatically mean job loss, and cited Office for Budget Responsibility figures are risk scenarios rather than observed layoffs | Three in four London jobs are highly exposed to generative AI, against a UK-wide highly exposed share of 60.9% | Share of jobs classed as highly exposed, defined in press coverage as roles where more than 50% of daily tasks could be performed by generative systems such as ChatGPT, Gemini or Claude |
Not at the aggregate level in this evidence set. Apollo's July 30 white paper found no statistically significant employment effect in its main model across 321 matched US occupations, locating the measured difference in wage growth instead. An Australian Department of Employment and Workplace Relations report, covered by ABC, found AI is not yet causing broad labour-market disruption based on occupational change since ChatGPT's November 2022 release, while noting early weakness in some highly exposed occupations including telemarketers. Israel's Employment Service reported more than 16,000 high-tech jobseekers, about three times the 2022 level, but CTech reported the researchers found correlation rather than proof of causation. Specific reductions exist, including monday.com's roughly 20% cut affecting about 620 people, but the company said the reduction was not designed to replace people with AI.
In relative wage growth and in the shape of premiums rather than in a uniform AI salary bump. Apollo estimated 6.7 percentage points slower real-wage growth in high-exposure occupations after 2023, with about 10.7% slower growth for the bottom wage quartile and roughly a 24.3% relative decline for service occupations; these are relative growth estimates, not findings that any worker took a pay cut of that size. Payscale's Ruth Thomas told Lets Data Science that AI fluency is already assumed in core tech and no longer moves the needle on pay, with 8 to 12% new-hire premiums appearing in functions such as Marketing Analyst III and IT Product Manager IV where the capability is still genuinely scarce. Indeed Hiring Lab separately found a 42% median hourly premium in data-center installation and maintenance postings, about $10 per hour, but that compares advertised pay across postings rather than realized earnings.
Google's ATLAS study, released July 23 from about 15 million de-identified interactions across the Gemini app, Search AI Mode and the Gemini API, found AI activity in 68% of occupations representing 90% of U.S. employment but in only about 21% of tasks within a typical job, with fewer than 10% of workplace interactions fully automating a task. Google notes that occupational reach and task coverage answer different questions, and that neither figure means 68% of workers actively use Gemini. OpenAI's Work at the Frontier report found 43.5% of occupation-specific messages concerned tasks historically associated with another occupation, with the highest crossover rates among customer-experience workers at 77% and designers at 75%. Both studies are bounded to their own products: Axios reports ATLAS excluded Gemini Enterprise and Workspace activity, and OpenAI says its sample should not be generalized to ChatGPT Enterprise users or the whole U.S. workforce.
Business Insider reported on July 31 that some technology employers are adding middle or final in-person rounds because candidates can use AI during virtual coding tests, citing Stanford economist Nick Bloom's comments to the Financial Times' Economics Show. Greenhouse's 2025 AI in Hiring report, which surveyed 4,136 job seekers, recruiters and hiring managers across the United States, United Kingdom, Ireland and Germany, found 39% of U.S. hiring managers were conducting more in-person interviews to verify candidates and 65% had caught applicants using AI deceptively; those are vendor survey results, not a census of employers. The countertrend is documented in the same reporting: Google piloted an approved AI assistant for a software-engineering interview round, and Cisco was described as shifting toward project-based exercises. A separate July 8 Business Insider report described interviews moving toward AI fluency, project reasoning and public proof of work on GitHub.
Employers report that it is, but the evidence is self-reported. A ResumeTemplates.com survey of 1,000 U.S. hiring managers, fielded by Pollfish in May 2026 at companies with at least 101 employees and carrying a margin of error of plus or minus 3.1 percentage points at 95% confidence, found 45% said their companies had restructured work so one senior employee using AI performs tasks previously handled by multiple entry-level graduates, and 55% said their employer had shifted at least part of its entry-level hiring budget to AI. Hiring plans were mixed rather than uniformly negative: 23% planned fewer or no 2026 graduate hires while 65% expected the same number or more. The report notes these responses describe employer sentiment, not audited payroll records or measured productivity comparisons. Toptal's Erik Stettler told Lets Data Science that a related risk is being spared too much foundational work, since AI removes the messy data preparation through which early-career analysts learned to spot outliers, reporting errors, selection effects and bias.
Several independent surveys point at the management layer rather than the tools. ManpowerGroup's Talent Solutions study, developed with Everest Group and based on 80 C-suite, CHRO and senior talent-acquisition leaders in the United States and United Kingdom, found only 3% of organizations considered their leaders highly prepared to manage AI-enabled ways of working, while 34% reported their largest productivity gains in redesigned AI-augmented roles versus 8% from fully automated roles. Indeed's July 28 survey of 1,301 U.S. respondents found 43% of managers felt unequipped to lead AI-native workers and 52% of workers said they lacked the AI training they needed. SANS's 2026 Cybersecurity Workforce Research Report, based on 947 respondents across six regions, found 74% of organizations seeing AI affect security-team size or role structures while only 38% provide comprehensive AI security training. These are bounded survey samples, several of them vendor-commissioned, so they describe reported practice rather than measured capability.
Latest coverage
Showing newest first with full-topic pagination
Atlassian reported on August 6 that fiscal Q4 revenue reached $1.8 billion, up 28% year over year, while customers added Jira and Confluence teams and users as they adopted AI. CEO Mike Cannon-Brookes then told Business Insider that developer hiring would keep growing, but the company’s platform results do not independently prove economy-wide software-job growth.
House Democrats introduced the AI Tax and Work Protection Act on August 6, proposing an excise tax on covered AI businesses to fund a federal jobs program. H.R. 10044 would use the greater of a token-value calculation or an AI-service revenue calculation, with rates linked to unemployment. The proposal was referred to two House committees and has not been enacted.
E.W. Scripps said it eliminated 268 jobs earlier this week, affecting about 6% of its workforce, as it restructures local television around streaming, centralized production, automation and AI. The company expects about $100 million in annual run-rate savings by year-end, while industry reporting says roughly a dozen smaller markets will begin the 24/7 local-streaming model.
Kookmin University said it launched the KMU AX Innovation Alliance at a July 24 campus meeting attended by representatives of 10 AI and software companies. The group discussed joint research, courses, internships and hiring pathways, but the university did not announce committed projects, funding, a timeline or firm membership terms.
Block reported second-quarter 2026 gross profit of $3.17 billion, up 25% year over year, and raised its full-year growth outlook to 21%. The company says agentic AI now helps write and review nearly all production-code changes; independent coverage reports 150% more code changes per engineer and 130 Square features shipped in the first half. Those throughput figures follow a workforce reduction of more than 40%, but do not by themselves measure product quality.
OpenAI OpCo and subsidiary Statsig agreed on August 4 to a $3.2 million Justice Department settlement over allegations that their PERM recruitment favored temporary visa holders and discouraged U.S. applicants. The resolution includes $1.2 million in civil penalties and a $2 million compensation fund; OpenAI told the Daily Caller that it disagreed with the findings but settled to continue its PERM program.
Two reports published July 28 say a Gartner survey of 110 HR chiefs found that 22% had at least one business leader stop hiring for entry-level roles because of AI automation. The same fourth-quarter 2025 survey found 95% of organizations had implemented AI in the prior year, but only one in five reported significant or transformational value, highlighting a gap between adoption and workforce planning.
Visa said July 28 that it would eliminate roughly 2,600 roles, about 7% of its workforce, with most reductions in technology and product operations, according to CNBC. CEO Ryan McInerney's memo linked AI to a broader shift in how work gets done, while a source told CNBC AI was significant but not the sole driver. India-based employees began receiving notices July 29, India Today reported.
Deel announced on August 3 that it acquired Clarity, an AI cybersecurity company focused on identity verification, deepfake detection, and fraud prevention. Financial terms were not disclosed; The Times of Israel reports that sources placed the deal at roughly $45 million to $50 million. Deel's announcement describes the acquisition as extending identity security from hiring through workforce access.
Kookmin University and the Korea International Cooperation Agency announced on Aug. 3 an AI doctoral fellowship for 30 Palestinian students at Kookmin's Graduate School of AI and Software. The program sits within a South Korean aid project running through 2031 and is designed to connect doctoral research with Palestinian digital-transformation needs.
On August 1, Elon Musk said SpaceX was seeking engineers and skilled-trades workers to build and operate AI supercomputer clusters on Earth and in space. Applicants were asked to email a resume and three bullet points showing exceptional ability. Business Insider linked the recruiting push to SpaceXAI's Memphis Colossus campus, while neither source disclosed roles, headcount, sites or construction timelines.
36Kr reported on July 31 that Ailing Zeng joined Bilibili to lead its AI video-generation business and report to CEO Chen Rui. Bilibili had not responded to the outlet's request for confirmation, and Zeng's personal profile still listed her at Anuttacon when retrieved on August 3, so the appointment remains unconfirmed by either party.
Indonesia and India are exploring joint AI research, talent exchanges, startup development and digital infrastructure projects following a July 31 session on Indonesia's AI roadmap. Deputy communications minister Nezar Patria said the countries' capabilities are complementary, while concrete programs, funding and delivery timelines remain to be defined.
Business Insider reported on July 31 that some technology employers are adding middle or final in-person interview rounds because candidates can use AI during virtual coding tests and interviews. Greenhouse's 2025 survey of 4,136 participants found that 39% of U.S. hiring managers were conducting more in-person interviews to verify candidates, showing the concern predates the latest comments.
U.S. Senators Mark Kelly, Jim Banks and Todd Young introduced the bipartisan AI DATA Act on June 10, 2026. S. 4742 would modernize three federal labor-market surveys, add quarterly AI-adoption questions to a Census Bureau business survey for 10 years, and require annual workforce-impact reports; it remains a proposal referred to the Senate labor committee, not enacted law.
Walmart's July 16, 2026 Jobs Spotlight report said the retailer is helping its 2.1 million associates build skills and confidence as AI changes work. The report follows April remarks from chief people officer Donna Morris that Walmart aims over several years to give every associate some level of AI skills; role-specific certifications were then available to 1.7 million workers in the United States and Canada.
Nigeria's Federal Government signed an agreement on July 30, 2026, to roll out an AI Teacher Capacity Development Programme for about 11,700 teachers in Federal Unity Colleges. The Federal Ministry of Education will implement the nationwide programme with ICEDT Consult Limited, targeting digital skills for lesson delivery, classroom assessment, and learning outcomes.
Apollo published a July 30 white paper estimating that real-wage growth in high-AI-exposure occupations was 6.7 percentage points slower after 2023 than in lower-exposure occupations, while the study detected no statistically significant employment effect. The difference-in-differences analysis matched 321 occupations and measured exposure using observed Claude use from Anthropic's Economic Index.
Shaw University announced on June 24, 2026, that SACSCOC approved it to confer doctoral degrees, clearing the way for its Divinity School to launch an EdD in Artificial Intelligence and Moral Agency. Shaw calls it the first program of its kind at an HBCU and will welcome a 12-student cohort in spring 2027. The practice-oriented program combines AI literacy, theological ethics, moral reasoning, social justice, and applied ethics.
Indeed Hiring Lab reported on July 14 that hourly data-center installation and maintenance postings carried a 42% median pay premium, about $10 more per hour, versus comparable non-data-center roles. PYMNTS highlighted the finding on July 29 as large projects pull skilled workers toward data-center construction and tighten labor competition elsewhere.
A federal judge in Michigan denied a University of Michigan student's request to litigate anonymously on July 28 in her lawsuit challenging discipline for alleged unauthorized AI use in course papers. Reason reports that the student received disciplinary probation and was ordered to write a 1,000-word reflective essay, then alleged violations of the ADA, Rehabilitation Act, and procedural due process rights.
OpenAI's July 27 Work at the Frontier report found that 43.5% of occupation-specific messages in its sample mapped to tasks historically associated with another occupation. The analysis covered more than 800,000 work-related messages from individual ChatGPT accounts of U.S. users whose roles were linked from Business account information; it did not measure productivity, output quality or employment effects.
Harvard Kennedy School unveiled its HKS 2036 strategy on June 3, setting a decade-long goal to train 600 public-sector technologists and 6,000 public leaders who can work across technology, governance, and ethics. The plan starts with a technology concentration in the master in public policy program and considers a separate technology-and-policy degree, tying AI fluency to broader public-service training.
Indeed research published July 28 found that 43% of surveyed managers felt poorly equipped or not equipped to lead AI-native workers, while 45% felt ready. The U.S. survey of 1,301 people also found 52% of workers lacked the AI training they needed, making management development and role-specific upskilling a practical constraint on AI adoption.
Toptal's Q2 2026 High-Skilled Job Report describes a split labor market: postings across all professions fell 3.6% quarter over quarter and 9% year over year, while demand for experienced remote and hybrid technology and professional services talent rose 5.4% and 8.9%. For data science specifically, full-time postings for experienced roles climbed 28% year over year with offered compensation flat and part-time compensation down 48%. In an email interview with Lets Data Science, Toptal Chief Economist Erik Stettler said employers are not paying for tenure but for judgment, and warned that early-career analysts risk losing the messy-data apprenticeship that taught them to catch bias before it reached a model.
Google released its first ATLAS study on July 23, analyzing about 15 million de-identified interactions across the Gemini app, AI Mode and Gemini API. It found AI activity across 68% of occupations representing 90% of U.S. employment, but only about 21% of tasks in a typical job. Manual and technical workers who used the tools were twice as likely to use multimodal AI.
ManpowerGroup's Talent Solutions unit released research on July 22 finding that only 3% of surveyed organizations considered their leaders highly prepared to manage AI-enabled work. The study also found that 17% reported advanced or transformational workforce readiness, while respondents linked their strongest productivity gains more often to redesigned AI-augmented roles than fully automated roles.
OpenAI said on July 27 that it will add 250 jobs in Dublin over the next two years and move its EU headquarters to the 88,000-square-foot Tropical Fruit Warehouse in late 2026. The company already employs more than 100 people in Ireland, so the plan would lift its local workforce above 350 if the hiring is completed.
monday.com announced on July 22 that it will reduce its global workforce by about 20%, affecting around 620 people, while reorganizing around its AI Work Platform. The company expects $45 million to $55 million in net restructuring charges, plans to keep hiring in selected strategic areas, and raised its 2026 non-GAAP operating-margin outlook to about 15%.
Bessemer Venture Partners' May 18 survey of 173 leaders at 113 portfolio companies found 49% were delivering more without adding headcount and 13% had slowed or paused hiring. Claude was used by 73% of respondents, while adoption varied sharply by function, suggesting teams are changing workflows and staffing plans unevenly rather than following a single AI playbook.
