State of AI Work 2026

Report AI · State of AI series · Q4 2026 edition · Quarterly

State of AI Work 2026

How AI changed who is cut, who is hired and what AI skills pay, 2024 to 2026: layoffs, reversals, entry-level hiring, AI use, the skills premium, productivity and the forecasts

Thank you for reading this post, don't forget to subscribe!

By the Report AI editorial team · Published · Next edition: Q1 2027
121,401 (Jan–Oct, to date)
US job cuts attributed to AI (Challenger), 5 Nov 2026 (Jan–Oct 2026) MEDIUM
617,189 (Jan–Oct, to date)
All announced US job cuts (Challenger), 5 Nov 2026 (Jan–Oct 2026) MEDIUM
19.7% (Jan–Oct, to date)
AI-attributed cuts as a share of all US announced cuts, 5 Nov 2026 (Jan–Oct 2026) DERIVED
193,226 (to date)
Cumulative US job cuts attributed to AI since 2023 (Challenger), 5 Nov 2026 (through Oct 2026) DERIVED
At a glance
  • Challenger, Gray & Christmas counted 121,401 US job cuts attributed to AI from January to October 2026, more than twice the 54,836 in all of 2025 and almost ten times the 12,742 in 2024.
  • AI-attributed US job cuts fell to 1,265 in October 2026, the lowest month of the year, and by Challenger’s 5 November 2026 report ‘market/economic conditions’ had overtaken AI as the leading stated reason for 2026 cuts.
  • Robert Half found in 2026 that 32% of US hiring managers who eliminated a role primarily because of AI had later rehired for it, typically in hybrid roles paying 20–35% more.
  • Stanford Digital Economy Lab’s August 2026 update found employment of 22-to-25-year-olds in AI-exposed US occupations 19% below where it would have been had it kept pace with less-exposed peers.
  • Indeed Hiring Lab data show US software-development job postings at 81.6% of their February 2020 level on 30 October 2026, up from 67.3% at the end of 2025, with senior roles leading the recovery.
  • Gallup found in May 2026 that 52% of US employees used AI at work at least a few times a year and 15% used it daily, against about 4% daily use in mid-2024.
  • PwC’s 2026 Global AI Jobs Barometer, published on 15 June 2026, found job advertisements requiring AI skills offered wages 62% higher on average than comparable roles without them.
Cite this report
Report AI, State of AI Work 2026, Q4 2026 edition, published 16 November 2026. https://report-ai.org/indexes/state-of-ai/state-of-ai-work/
Download the data (CSV) · 27 metrics with source, date and confidence for every value. Free to quote with attribution and a link.
In this report
  1. 1. Layoffs — US job cuts attributed to AI rose almost tenfold in two years, then fell to their lowest monthly level of 2026 in October
  2. 2. Reversals — About a third of US hiring managers who cut a role for AI have already rehired for it, usually at higher pay
  3. 3. Entry level — Software hiring recovered in 2026, but the gap for young workers in AI-exposed jobs widened to 19%
  4. 4. Use, demand and pay — AI moved into a majority of jobs in 2026, demand for AI skills nearly tripled in postings, and the advertised pay premium kept rising
  5. 5. Exposure and productivity — Exposure is highest in translation, administrative and legal work, but measured productivity gains across whole jobs remain small
  6. 6. Forecasts — The leading forecast still expects AI-era job creation to exceed displacement by 78 million worldwide by 2030, but it is a forecast built on employer expectations
  7. Comparison
  8. The numbers, 2024–2026
  9. Outlook: what to watch in 2027
  10. FAQ · About the research
1. Layoffs

US job cuts attributed to AI rose almost tenfold in two years, then fell to their lowest monthly level of 2026 in October

The most quoted measure of AI’s effect on jobs is a count of announcements. Challenger, Gray & Christmas, an outplacement firm, records the job cuts US employers announce each month and the reason they give. On that measure, AI went from a rounding error to a headline in two years. Challenger counted 12,742 US job cuts attributed to AI in 2024, 54,836 in 2025 and 121,401 from January to October 2026. As a share of all announced cuts, by Report AI’s division of Challenger’s totals, that is about 1.7% in 2024, 4.5% in 2025 and 19.7% in the first ten months of 2026. Since tracking began in 2023, employers have attributed roughly 193,000 announced US job cuts to AI, by Report AI’s sum.

The 2026 total was built in a few months. Challenger’s cumulative count reached 49,135 by the end of April and 87,714 by the end of May, so May alone accounted for about 38,600 AI-attributed cuts by Report AI’s subtraction, roughly 40% of all cuts announced that month. In March AI became the single most-cited reason for US job cuts, at 25% of that month’s total. The monthly count then fell steadily, by the same subtraction: about 14,000 in June, 11,000 in July, under 4,000 in each of August and September, and 1,265 in October, the lowest month of the year. By Challenger’s 5 November 2026 report, ‘market/economic conditions’ (130,531 cuts) had overtaken AI as the leading stated reason for the year to date.

Exhibit 1
AI-attributed US job cuts in January–October 2026 were already more than twice the full-year 2025 total
US announced job cuts for which employers cited artificial intelligence, by calendar year; 2026 is January–October
025,00050,00075,000100,000125,000job cuts2024AI-attributed job cuts, 2024: 12,742 job cuts12,7422025AI-attributed job cuts, 2025: 54,836 job cuts54,8362026 (Jan–Oct)AI-attributed job cuts, 2026 (Jan–Oct): 121,401 job cuts121,401
View the data
AI-attributed job cuts
202412,742
202554,836
2026 (Jan–Oct)121,401
Source: Challenger, Gray & Christmas monthly job-cut reports (via secondary reporting); Report AI analysis MEDIUM
Notes: 2024 (12,742) is from a single secondary summary, MEDIUM; consistent with Challenger’s cumulative 71,825 since 2023. 2025 (54,836) and 2026 (121,401, to 31 Oct, reported 5 Nov 2026) MEDIUM. 2026 is a partial year. Share of all announced cuts (DERIVED): 1.7%, 4.5%, 19.7%.REPORTAI report-ai.org

The fall in AI-attributed cuts happened alongside a fall in layoffs generally. All announced US cuts in January–October 2026 totalled 617,189, down 44% on the same period of 2025, and October’s 43,994 was 71% below October 2025. A lower AI count in a quieter layoff market is therefore not, on its own, evidence that AI-driven restructuring has ended. It is consistent with several readings: large employers that planned AI-justified cuts announced them early in the year; some employers have become more cautious about naming AI after public reversals; or the first wave of restructuring has simply run its course. One month cannot distinguish between these. September’s 3,961 AI-attributed cuts already ranked only fifth among stated reasons that month, so AI had dropped down the list before October’s low.

The count should be read as what it is. It records the reason employers choose to give, which can overstate AI’s role, because ‘AI’ reads as modernisation rather than weak demand, and understate it, because some cuts are filed under headings such as ‘technological update’ or ‘restructuring’. It covers announcements, not completed separations, and it is skewed towards large employers that announce cuts publicly. As a measure of direction and of how employers talk about AI it is useful; as a count of jobs destroyed by AI it is neither a floor nor a ceiling. Where we quote AI’s share of all cuts, it is Report AI’s division of Challenger’s own totals, labelled DERIVED, and the annual counts are rated MEDIUM because Challenger’s releases could not be opened directly from our research environment.

Exhibit 2
Nine-tenths of 2026’s AI-attributed cuts had been announced by the end of July; the count barely moved after that
Cumulative US announced job cuts attributed to AI since 1 January 2026, at month-end, where Challenger or reporting on it gave a figure
025,00050,00075,000100,000125,000job cuts (cumulative)OctSepJulJunMayAprAI-attributed cuts, 2026 YTD, Apr: 49,135 job cuts (cumulative)AI-attributed cuts, 2026 YTD, May: 87,714 job cuts (cumulative)AI-attributed cuts, 2026 YTD, Jun: 101,743 job cuts (cumulative)AI-attributed cuts, 2026 YTD, Jul: 112,713 job cuts (cumulative)AI-attributed cuts, 2026 YTD, Sep: 120,136 job cuts (cumulative)AI-attributed cuts, 2026 YTD, Oct: 121,401 job cuts (cumulative)121,40149,135
View the data
AI-attributed cuts, 2026 YTD
Apr49,135
May87,714
Jun101,743
Jul112,713
Sep120,136
Oct121,401
Source: Challenger, Gray & Christmas monthly job-cut reports (via secondary reporting); Report AI analysis MEDIUM
Notes: Plotted only months with a published cumulative figure: Jan–Apr 49,135; Jan–May 87,714 (secondary reporting of Challenger’s May report); H1 ~101,743 (Challenger via Report AI page 88); Jan–Jul 112,713 (secondary reporting); Jan–Sep 120,136 (Challenger, 1 Oct 2026); Jan–Oct 121,401 (Challenger, 5 Nov 2026). Monthly values quoted in the text are DERIVED by subtraction (May ≈38,579; Jun ≈14,029; Jul ≈10,970; Aug ≈3,462; Sep 3,961 reported; Oct 1,265 reported). 112,713 ÷ 121,401 = 93% of the Jan–Oct total (DERIVED).REPORTAI report-ai.org
How Challenger counts AI job cuts

Challenger, Gray & Christmas compiles job cuts that US employers announce publicly or in filings and classifies each by the reason the employer gives. ‘Artificial intelligence’ has been a category since 2023. The count is of planned cuts at announcement, not separations, and it leans towards large employers. A separate category, ‘technological update’, can include AI-related cuts that are not described as such. Because the reason is self-reported, the count measures what employers say about AI as much as what AI does. Challenger’s primary releases could not be opened from our research environment; the figures here come from consistent secondary reporting and are rated MEDIUM.

2. Reversals

About a third of US hiring managers who cut a role for AI have already rehired for it, usually at higher pay

The second half of the AI-layoff story is the cost of undoing it. Robert Half’s 2026 survey of about 2,000 US hiring managers found that 32% of those whose organisation eliminated a role primarily because of AI had later rehired for that role or one like it. The rehired positions were rarely the old jobs restored: they typically came back as hybrid roles combining domain knowledge with AI-tool fluency, and paid 20–35% more than the roles they replaced. Among those who rehired, 40% said the role required institutional knowledge or context that AI could not replace, and 38% said AI needed more human oversight and quality control than expected.

Other survey data point the same way. Forrester reports that 55% of employers regret AI-related layoffs. Survey data on organisations that reversed found that 30.9% spent more on rehiring than automation had saved and a further 42.4% roughly broke even, so about 73% ended no better off than if they had never cut, by Report AI’s sum of the two groups. In customer service, only 20% of service leaders say they have actually reduced agent numbers because of AI, and 95% say they plan to keep human agents. These are vendor surveys from firms that sell into the hiring market, and they are rated MEDIUM; they are consistent with the company cases but are not national statistics.

Exhibit 3
Survey and company evidence agree that AI-justified cuts are frequently undone, though most of the survey data is vendor research
Selected findings on reversal and rehiring after AI-attributed job cuts, 2024–2026
FindingFigureSourceConfidence
US hiring managers who cut a role for AI and later rehired for it32%Robert Half, 2026MEDIUM
Pay premium on rehired hybrid roles20–35%Robert Half, 2026MEDIUM
Employers who regret AI-related layoffs55%Forrester, 2026MEDIUM
Reversing organisations no better off than before cutting~73%Report AI from survey dataDERIVED
Service leaders who actually cut agents because of AI20%Survey data, 2026MEDIUM
AI-attributed layoffs forecast to be reversed by end-2026~50%Forrester (forecast)MEDIUM
Klarna: AI agent-equivalents, Feb 2024 → Nov 2025~700 → 853KlarnaHIGH
Meta, 2026 vs 2025: code changes / shipped features / major incidents+220% / +36% / +40%Reuters, 26 Aug 2026HIGH
Source: Robert Half; Forrester; Forrester-linked survey data; customer-service leader surveys; Klarna disclosures; Reuters (26 Aug 2026); Report AI analysis MEDIUM
Notes: Denominators differ row to row, so the percentages cannot be compared with each other. The ~73% is DERIVED (30.9 + 42.4) and assumes the two groups do not overlap. Forrester’s ~50% is a forecast, not an outcome. Meta figures are from a June 2026 internal post by CTO Andrew Bosworth as reported by Reuters (HIGH).REPORTAI report-ai.org

The named cases are better documented. Klarna said in February 2024 that its AI assistant was doing the work of about 700 customer-service agents; in May 2025 its chief executive said cost had been ‘too predominant’ a factor and the company began rehiring humans for complex cases, while the AI went on to do the work of 853 employees by November 2025. Meta abandoned the second planned wave of its Project OT cuts in 2026 after AI-assisted code changes rose 220% but shipped features rose only 36% and major incidents rose 40%, according to an internal post reported by Reuters on 26 August 2026. IBM, Duolingo and Salesforce have also scaled back or rehired against AI-replacement decisions.

The pattern is re-scoping, not abandonment. Klarna kept its automation for routine queries; IBM kept the roughly 94% of routine HR tasks it automated and found the remaining 6% resistant. What was reversed in each case was the assumption that handling a workload means replacing the people who handled its hardest part. Forrester’s forecast that about half of AI-attributed layoffs will be reversed in some form by the end of 2026 resolves in January 2027, and its outcome depends heavily on what counts as a reversal, a point our predictions scoreboard records. Duolingo’s chief executive described the same residual problem in a different function when reversing its ‘AI-first’ memo: on edge cases, ‘you end up spending more effort on that than the time you saved’.

Why the reversal data is rated MEDIUM

Most aggregate figures on regret and rehiring come from Forrester, Robert Half and similar firms, which sell research, recruitment or workforce services into the market they describe. Their surveys are not random samples of employers, question wording varies, and some figures circulate through secondary reporting with slightly different values (Robert Half’s rehire share appears as both 29% and 32%). We publish them because they agree in direction with the named company cases, where an executive is on the record. They should not be quoted as national statistics, and the 55% regret figure measures sentiment, not completed reversals.

3. Entry level

Software hiring recovered in 2026, but the gap for young workers in AI-exposed jobs widened to 19%

The clearest sign of AI in employment data is at the bottom of the career ladder. Stanford Digital Economy Lab researchers led by Erik Brynjolfsson, using payroll records from ADP covering millions of US workers, reported in August 2025 a 13% relative decline since late 2022 in employment of 22-to-25-year-olds in the most AI-exposed occupations, revised to 16% in November 2025 after controlling for firm-level shocks. Their August 2026 update, with data through June 2026, puts employment of that group 19% below where it would be had it kept pace with less-exposed peers. Older workers in the same occupations held steady or grew. The authors describe the result as descriptive, and find no evidence of economy-wide displacement.

Hiring data at large technology firms tells a similar story. SignalFire reported that new graduates made up about 7% of hires at large tech firms in 2024, roughly half the pre-pandemic level, and its 2026 analysis puts new-graduate hiring at 12 large firms about 65% below 2019. Industry posting analyses compiled on our software-developer page put entry-level developer postings 67% lower over 2022–2026, and 54% of companies say they stopped hiring junior developers because of AI. These are vendor figures, rated MEDIUM, and the two SignalFire numbers are different report vintages rather than one series. SignalFire also reports that engineering new-graduate hiring at those firms fell far less than design hiring, about 11% against about 48% below 2019, so the squeeze is uneven even within technology.

Exhibit 4
US software-development postings were flat at about two-thirds of their 2020 level for two years, then recovered sharply in 2026
Indeed job postings index, US, seasonally adjusted, 1 Feb 2020 = 100; software development vs all postings; quarter-ends Dec 2023–Sep 2026 and 30 Oct 2026
Software developmentAll US postings0255075100125index (Feb 2020 = 100)30 Oct 2026Jun 2026Dec 2025Jun 2025Dec 2024Jun 2024Dec 2023Software development, Dec 2023: 72.6 index (Feb 2020 = 100)Software development, Jun 2024: 70.1 index (Feb 2020 = 100)Software development, Dec 2024: 67.5 index (Feb 2020 = 100)Software development, Mar 2025: 62.5 index (Feb 2020 = 100)Software development, Jun 2025: 65.5 index (Feb 2020 = 100)Software development, Sep 2025: 64.3 index (Feb 2020 = 100)Software development, Dec 2025: 67.3 index (Feb 2020 = 100)Software development, Mar 2026: 72.9 index (Feb 2020 = 100)Software development, Jun 2026: 73.4 index (Feb 2020 = 100)Software development, Sep 2026: 78.3 index (Feb 2020 = 100)Software development, 30 Oct 2026: 81.6 index (Feb 2020 = 100)81.6Software development72.6All US postings, Dec 2023: 121 index (Feb 2020 = 100)All US postings, Jun 2024: 113 index (Feb 2020 = 100)All US postings, Dec 2024: 111 index (Feb 2020 = 100)All US postings, Mar 2025: 107 index (Feb 2020 = 100)All US postings, Jun 2025: 105 index (Feb 2020 = 100)All US postings, Sep 2025: 101 index (Feb 2020 = 100)All US postings, Dec 2025: 103 index (Feb 2020 = 100)All US postings, Mar 2026: 102 index (Feb 2020 = 100)All US postings, Jun 2026: 101 index (Feb 2020 = 100)All US postings, Sep 2026: 104 index (Feb 2020 = 100)All US postings, 30 Oct 2026: 104 index (Feb 2020 = 100)104All US postings121
View the data
Software developmentAll US postings
Dec 202372.6121
Jun 202470.1113
Dec 202467.5111
Mar 202562.5107
Jun 202565.5105
Sep 202564.3101
Dec 202567.3103
Mar 202672.9102
Jun 202673.4101
Sep 202678.3104
30 Oct 202681.6104
Source: Indeed Hiring Lab, Job Postings Index and AI Tracker datasets (github.com/hiring-lab, read 11 Nov 2026); Report AI analysis HIGH
Notes: Seven-day trailing averages read from Hiring Lab’s published data (commit of 4 Nov 2026), HIGH. Indeed revises historical values when it re-estimates seasonal factors; values may differ from earlier Hiring Lab posts. Postings measure demand advertised on Indeed, not hires.REPORTAI report-ai.org

The overall software market moved the other way in 2026. Indeed’s index of US software-development postings, set to 100 on 1 February 2020, sat near 67 at the end of both 2024 and 2025, with a low of 62.5 in March 2025. It then rose to 81.6 on 30 October 2026, still 18.4% below its pre-pandemic level but up 21% since the end of 2025. Hiring Lab reports that senior roles accounted for 71% of the gain in the year to May 2026. Total US postings were 3.8% above their February 2020 level at the same date. AI did not stop developer hiring; the recovery is concentrated in experienced roles.

Graduate hiring beyond tech is not collapsing. The National Association of Colleges and Employers’ spring 2026 update projected that US employers would hire 5.6% more graduates from the class of 2026 than from the class of 2025, after projecting only 0.6% growth for the class of 2025. That projection uses a median rather than the mean NACE used in its autumn forecast, so the two are not strictly comparable. The combined reading is narrower than the headlines: entry into AI-exposed occupations is getting harder, while entry-level hiring as a whole is roughly flat to modestly up. Customer support shows the same pattern by a different route: fewer tier-1 roles means fewer people learning a product well enough to become the escalation specialists that the same organisations now depend on.

Exhibit 5
Every early-career indicator in AI-exposed work points down, while broad graduate hiring projections point up
Entry-level and graduate hiring indicators, US, 2024–2026
Indicator202420252026
Stanford: employment gap, ages 22–25, AI-exposed jobs—16% (Nov)19% (Aug)
SignalFire: Big Tech new-grad hiring vs 2019about −50%—about −65%
Entry-level developer postings since 2022——−67%
Companies that stopped hiring junior developers for AI——54%
NACE: projected graduate hiring change—+0.6%+5.6%
Indeed: software-development postings vs Feb 2020−32.5%−32.7%−18.4% (Oct)
Source: Stanford Digital Economy Lab (Brynjolfsson, Chandar & Chen); SignalFire; industry posting analyses; NACE; Report AI analysis MEDIUM
Notes: Measures are not comparable with each other. Stanford figures are relative employment gaps (13% Aug 2025 → 16% Nov 2025 revision → 19% Aug 2026 update; RESTATED, descriptive). SignalFire covers 12 large tech firms; the −50% and −65% are different report vintages. NACE figures are employer projections (class of 2026 on a median basis).REPORTAI report-ai.org
Postings are not hires

Job-posting indices measure how much hiring employers advertise, not how many people are hired. They lead actual hiring, can overstate turning points in both directions, and depend on one platform’s share of the market. Indeed’s software-development index counts postings classified by normalised job title, seasonally adjusted, against a 1 February 2020 baseline that was itself near a tech-hiring peak. Payroll-based studies such as Stanford’s Canaries series measure employment directly but cannot isolate AI from interest rates, post-pandemic over-hiring or other shocks, which is why the authors call their results descriptive.

4. Use, demand and pay

AI moved into a majority of jobs in 2026, demand for AI skills nearly tripled in postings, and the advertised pay premium kept rising

Using AI at work became a majority habit in 2026, though a light one for most. Gallup found in May 2026 that 52% of US employees used AI in their role at least a few times a year, up from 40% in the second quarter of 2025, and that 30% used it a few times a week or more. Daily use reached 15%, against about 4% in the second quarter of 2024. Pew Research Center, asking a different question, found that 21% of US workers said at least some of their work was done with AI in September 2025, up from 16% in October 2024. Gallup changed its question in the third quarter of 2025 by adding a ‘don’t know’ option, so later readings are not strictly comparable with earlier ones.

Employers are asking for AI in job descriptions much more often. Indeed Hiring Lab data show that 2.71% of US postings mentioned AI-related terms at the end of 2024, 4.44% at the end of 2025 and 7.34% on 30 September 2026, about 2.7 times in 21 months. In the UK the share rose from 3.44% to 9.92% over the same period. Lightcast, using a narrower definition of postings that require AI skills, put the US share at 1.8% for 2024 and about 2.5–2.6% for 2025 in Stanford’s AI Index. The level depends on the definition; every series shows the same direction. In Canada the Indeed share reached 18.97% by September 2026, partly because Ontario began requiring employers to disclose AI use in hiring from January 2026, which is why cross-country levels should not be ranked against each other.

Exhibit 6
The share of job postings mentioning AI has nearly tripled in the US and the UK since the end of 2024
Share of all job postings on Indeed mentioning AI-related terms (e.g. ‘machine learning’, ‘artificial intelligence’), %, seven-day trailing average, quarter-ends Dec 2023–Sep 2026
United StatesUnited Kingdom0246810% of postingsSep 2026Mar 2026Jun 2025Dec 2024Jun 2024Dec 2023United States, Dec 2023: 1.94 % of postingsUnited States, Jun 2024: 2.23 % of postingsUnited States, Dec 2024: 2.71 % of postingsUnited States, Jun 2025: 3.1 % of postingsUnited States, Dec 2025: 4.44 % of postingsUnited States, Mar 2026: 5.44 % of postingsUnited States, Jun 2026: 5.99 % of postingsUnited States, Sep 2026: 7.34 % of postings7.34United States1.94United Kingdom, Dec 2023: 2.3 % of postingsUnited Kingdom, Jun 2024: 2.48 % of postingsUnited Kingdom, Dec 2024: 3.44 % of postingsUnited Kingdom, Jun 2025: 4.46 % of postingsUnited Kingdom, Dec 2025: 7.12 % of postingsUnited Kingdom, Mar 2026: 8.13 % of postingsUnited Kingdom, Jun 2026: 9.5 % of postingsUnited Kingdom, Sep 2026: 9.92 % of postings9.92United Kingdom2.3
View the data
United StatesUnited Kingdom
Dec 20231.942.3
Jun 20242.232.48
Dec 20242.713.44
Jun 20253.14.46
Dec 20254.447.12
Mar 20265.448.13
Jun 20265.999.5
Sep 20267.349.92
Source: Indeed Hiring Lab, Job Postings Index and AI Tracker datasets (github.com/hiring-lab, read 11 Nov 2026); Report AI analysis HIGH
Notes: Read from Hiring Lab’s AI Tracker data (commit of 8 Oct 2026), HIGH. Values are revised over time: Hiring Lab wrote in January 2026 that the US tracker reached 4.2% at end-2025; the current dataset shows 4.44% (RESTATED). Not comparable with Lightcast’s ‘postings requiring AI skills’ (US 1.8% in 2024; ~2.5–2.6% in 2025, Stanford AI Index).REPORTAI report-ai.org

Advertised pay for AI skills has risen in each edition of PwC’s Global AI Jobs Barometer. Across roughly a billion job advertisements in many countries, PwC found a wage premium of 25% for jobs requiring AI skills in 2023 advertisements, 56% in 2024 advertisements (later restated as 57%) and 62% in the 2026 edition published on 15 June 2026, which analyses the previous year’s advertisements. The premium ranged from 118% in consumer markets to 16% in government, and jobs requiring AI skills grew about 69% against about 9% for all jobs. PwC describes the labour market as splitting into ‘professionalised’ roles, which add new skills about twice as fast, and ‘democratised’ roles in which AI use becomes routine; the premium is largest in the first group.

The premium is an advertised-wage comparison between similar roles with and without an AI-skill requirement. It is not a measured pay rise for workers who learn AI, and it may partly reflect which employers and roles ask for AI skills. PwC’s figures are averages across job advertisements in many countries, not US payroll data, and should not be read against US wage statistics. Gallup’s industry split shows a similar concentration in use: in the third quarter of 2025, 76% of technology workers used AI at least a few times a year, against 33% in retail, 37% in health care and 38% in manufacturing. Rising demand and pay for AI skills sit alongside falling early-career employment in AI-exposed jobs; they describe different workers.

Two ways to count AI job postings

Indeed’s AI Tracker counts the share of postings that mention any of a basket of AI-related terms anywhere in the title or description, including postings where AI is a tool the worker will use. Lightcast, whose data feeds Stanford’s AI Index, counts postings that require at least one AI-related skill. The first is broader and higher (7.34% in the US in September 2026); the second is narrower (about 2.5% for 2025). Neither is wrong. Compare readings within a series, never across them, and note that Indeed revises recent values as its data are re-estimated.

5. Exposure and productivity

Exposure is highest in translation, administrative and legal work, but measured productivity gains across whole jobs remain small

Exposure estimates rank occupations by how much of their work current AI could do. The IMF estimated in January 2024 that about 40% of global employment is exposed to AI, rising to about 60% in advanced economies, where more work is cognitive. At the occupational level, the studies compiled on our by-occupation page put interpreters and translators above 98% automation exposure, market research at 53% of analyst tasks, office and administrative support at a 46% task-automation share and legal work at 44%. These come from different studies with different methods; they are a ranking, not a set of comparable measurements. Translation is the clearest measured case: machine translation crossed the threshold of commercial acceptability for routine content, and surviving demand is concentrated in legal, medical and certified work where accuracy carries liability.

The least exposed work is defined less by industry than by the content of the job. Roles where judgement under ambiguity, personal accountability, physical presence or human trust is the product sit near the bottom of every exposure ranking. Payroll data reflect the split: Stanford’s August 2026 update reports that employment of 22-to-25-year-olds in the three least-exposed occupational quintiles grew about 10% over the period studied, while it fell about 11% in the most exposed. Exposure is also a poor predictor of headcount. Office and administrative support has the highest task share yet not the highest observed job loss, because the remaining 54% of tasks still needs accountable people. IBM’s HR automation is the standard example: it automated about 94% of routine tasks and then moved to triple US entry-level hiring because the residual required judgement.

Measured productivity effects are smaller than exposure estimates imply. The St. Louis Fed’s Real-Time Population Survey put time saved by generative AI at about 1.4% of all US work hours in November 2024, rising to about 2% by the fourth quarter of 2025 and about 2.2% in the first quarter of 2026; workers who use AI report saving about 5.4% of their own hours. These are self-reports, and saved time raises output only if it is redeployed. Controlled studies are mixed: a 2023 field study of customer-support agents found 14% more issues resolved per hour, while METR’s randomised trial, published in July 2025, found experienced developers took 19% longer with early-2025 AI tools.

Exhibit 7
The most exposed occupations are those whose tasks are routine, checkable and low-liability; the least exposed carry judgement or accountability
AI exposure estimates by occupation and what has measurably happened, 2024–2026
Occupation or groupExposure estimateWhat has measurably happened
Interpreters and translators98%+ automation exposureRoutine translation largely automated; certified and legal work retained
Market research analysts53% of analyst tasksData gathering and first-pass synthesis absorbed
Office and administrative support46% task share (highest)Scheduling and routing absorbed; no proportional job loss
Legal support and paralegals44% task shareDocument review time cut sharply; advice and appearance unaffected
Content and copywritingFreelance postings −33% since 2022Clearest displacement outside translation
Customer support (tier 1)Median deflection 41.2%Most automated and most reversed function
Software engineeringEntry-level postings −67% (2022–26)Review overtook writing; senior hiring recovering
Least-exposed quintiles, ages 22–25Lowest exposureEmployment about +10% vs about −11% in most exposed (Stanford)
All global employment (IMF)~40% exposed (~60% advanced economies)Exposure, not job loss
Source: IMF (Jan 2024); studies compiled on Report AI’s AI-exposure-by-occupation page (SHRM, Josh Bersin, Bloomberry, OECD and others); Stanford Digital Economy Lab (Aug 2026); Report AI analysis MEDIUM
Notes: Exposure measures differ by row (whole-role exposure, task share, projected reduction) and are not comparable; read as a ranking. All MEDIUM; several were gathered via search summaries. The least-exposed row is a relative employment change for workers aged 22–25 by exposure quintile (Stanford, Aug 2026 update, secondary reporting).REPORTAI report-ai.org

Firms report little effect so far. In the NBER study ‘Firm Data on AI’, published in February 2026 and based on surveys of about 6,000 executives in the US, UK, Germany and Australia, more than 90% reported no effect of AI on their firm’s employment over the previous three years and 89% reported none on labour productivity. The same executives expect a 1.4% productivity gain and a 0.7% employment decline over the next three years. Inside software teams, the gains in output are real but partly offset by verification: survey data put weekly time spent reviewing AI-generated code at 11.4 hours in the first quarter of 2026, against 9.8 hours writing new code.

Customer support, the first function automated at scale, shows the same gap between capability and replacement. Across enterprise programmes in 2026, the median share of tier-1 contacts resolved by AI without a human was 41.2%, with the top quartile at 58.7% and the bottom at 22.4%, well below the 70–90% that vendors advertise. Simple, lookup-based requests such as refunds and password resets deflect at 65–80%; nuanced complaints rarely exceed 25%. Re-contact rates were 11.3% on AI-resolved tickets against 8.7% for human-resolved ones, so part of the measured saving reappears later as repeat work. Satisfaction was highest, at about 89%, where AI triaged and humans handled escalations, against about 74% for AI alone. These are 2026 vendor benchmarks, rated MEDIUM, and deflection is defined differently by different vendors.

Exhibit 8
Self-reported time savings from generative AI rose to about 2% of US work hours, while most executives report no productivity effect yet
Measured and reported productivity effects of AI, selected studies, 2023–2026
Study or measureFindingPeriodConfidence
St. Louis Fed: share of all US work hours saved1.4% → ~2.0% → ~2.2%Nov 2024 → Q4 2025 → Q1 2026MEDIUM
Customer-support agents, field study+14% issues resolved per hourData 2020–21; published 2023MEDIUM
METR randomised trial, experienced developersTasks took 19% longer with AIFeb–Jun 2025MEDIUM
NBER executive survey: no productivity effect, past 3 years89% of executivesFielded late 2025–early 2026MEDIUM
NBER executive survey: expected effect, next 3 yearsProductivity +1.4%; employment −0.7%ExpectationMEDIUM
Developers: hours reviewing AI code vs writing code11.4h vs 9.8h per weekQ1 2026MEDIUM
Meta: code changes vs shipped features+220% vs +36%2026 vs 2025HIGH
Source: Federal Reserve Bank of St. Louis (Bick, Blandin & Deming; FRED series RPSGENAITSALL); METR (10 Jul 2025); Brynjolfsson, Li & Raymond (NBER 2023; QJE 2025); NBER w34836 ‘Firm Data on AI’ (Feb 2026); developer survey data and Reuters as compiled by Report AI MEDIUM
Notes: St. Louis Fed 2025 and 2026 values were read from a secondary summary of the FRED series (MEDIUM); the 2024 working-paper estimate may be constructed differently. The customer-support study measured 2020–21 data. METR is a small randomised trial on mature repositories; a later METR study reportedly did not replicate the slowdown (unverified). Faros-type delivery metrics are associations, not causal effects.REPORTAI report-ai.org
Exposure is not job loss

An exposure estimate measures how much of an occupation’s work current AI could perform or assist, not how many jobs will disappear. A job is a bundle of tasks; when part of the bundle is automated, the role usually changes shape and loses some headcount rather than vanishing, and the surviving work concentrates in judgement, exceptions and accountability. Task-automation share, whole-role exposure and projected headcount reduction are three different measures, and studies use different methods and time horizons. Treat any single exposure percentage as a position in a ranking, not as a forecast of jobs lost.

6. Forecasts

The leading forecast still expects AI-era job creation to exceed displacement by 78 million worldwide by 2030, but it is a forecast built on employer expectations

The most cited forecast of AI’s effect on jobs is the World Economic Forum’s Future of Jobs Report 2025, published in January 2025 and based on a survey of employers covering more than 14 million workers. It projects that structural change will create 170 million jobs and displace 92 million worldwide between 2025 and 2030, a net gain of 78 million, with 22% of today’s jobs churned and 39% of workers’ core skills changing. The drivers include technology, the green transition, demographics and economic conditions; the 92 million is not a forecast of jobs lost to AI alone. The same survey found 41% of employers planning workforce reductions where AI can automate tasks, alongside plans to hire for new AI and data skills, which is why the report reads as churn rather than simple loss.

The forecast has changed sign between editions. The 2023 edition, based on 803 companies employing 11.3 million people, projected 69 million jobs created and 83 million eliminated between 2023 and 2027, a net decline of 14 million. The 2025 edition covers a different horizon and sample, so the two are not directly comparable, but the change is a reminder that these are survey-based projections that move with employer sentiment. No newer edition had been published by November 2026; the 2025 projection is tracked on our predictions scoreboard. Forecasts of this kind are hard to score: by 2030 the comparison will largely be with the WEF’s own later editions, because national statistics cannot separate AI from the other drivers the survey covers.

Exhibit 9
The WEF’s forecast moved from a net loss of 14 million jobs to a net gain of 78 million between its 2023 and 2025 editions
World Economic Forum Future of Jobs projections: jobs created, displaced and net change, millions, by report edition (forecasts, all structural drivers)
2023 ed. (to 2027)2025 ed. (to 2030)-50050100150200M jobsCreated2023 ed. (to 2027), Created: 69 M jobs692025 ed. (to 2030), Created: 170 M jobs170Displaced2023 ed. (to 2027), Displaced: 83 M jobs832025 ed. (to 2030), Displaced: 92 M jobs92Net2023 ed. (to 2027), Net: -14 M jobs-142025 ed. (to 2030), Net: 78 M jobs78
View the data
2023 ed. (to 2027)2025 ed. (to 2030)
Created69170
Displaced8392
Net-1478
Source: World Economic Forum, Future of Jobs Report 2023 (May 2023) and 2025 (Jan 2025); Report AI analysis MEDIUM
Notes: FORECASTS, not outcomes. Horizons differ (2023 edition: 2023–2027; 2025 edition: 2025–2030) and samples differ (803 companies / 11.3M workers vs employers covering 14M+ workers), so the editions are not directly comparable. 2023-edition figures (69M, 83M, −14M) from WEF’s 2023 release, MEDIUM. Covers all drivers, not AI alone. Never compare with Challenger’s US layoff counts.REPORTAI report-ai.org

Set against the forecasts, the measured US evidence is small in aggregate and concentrated in places. AI-attributed announced cuts totalled about 193,000 from 2023 to October 2026, by Report AI’s sum of Challenger figures, and were 4.5% of all announced US cuts in 2025, the last full year. Over the same period, the share of US Indeed postings mentioning AI rose from 2.71% to 7.34%, and US software-development postings recovered to 18.4% below their 2020 level, so the measured US record shows reallocation within the labour market rather than a contraction of it. Executives surveyed for NBER expect a 0.7% employment decline at their own firms over the next three years, while their employees expect a 0.5% rise. Neither the WEF’s net gain nor the executives’ modest decline can yet be checked against outcomes, and neither should be quoted as a measurement.

The two kinds of evidence answer different questions. Forecasts describe what employers expect across all jobs and all causes; layoff counts, payroll studies and postings describe what has happened, mostly in the US, in specific occupations. The measured record so far supports a narrower claim than either the optimistic or the pessimistic forecast: AI is changing who gets hired and for what, faster than it is changing how many people are employed. Any sentence that sets a forecast of global job creation against a count of US layoffs compares unlike things. The more useful question for 2027 is which of the measured indicators in this report moves first, and in which direction.

Why forecasts disagree

Job forecasts differ because they measure different things. The WEF asks employers what they expect across all structural trends; the IMF estimates exposure, not loss; executive surveys such as NBER’s ask about own-firm effects over three years; vendor forecasts such as Forrester’s often concern a narrow event like rehiring. Horizons, geographies and definitions of ‘AI’ differ too. A forecast that changes sign between editions is not necessarily wrong, but it is evidence of how sensitive these projections are to the moment they are taken. We log forecasts with their resolution dates so they can be scored when they come due.

Comparison

Between 2024 and 2026, AI-attributed cuts and AI demand in job postings rose almost tenfold and the share of postings mentioning AI nearly tripled, while early-career employment in exposed jobs kept falling behind

Read year over year, the indicators in this report move together in direction but not in scale. AI-attributed US job cuts rose from 12,742 in 2024 to 54,836 in 2025 and 121,401 in January–October 2026, from about 1.7% to 19.7% of all announced cuts. Over the same period the share of US Indeed postings mentioning AI rose from 2.71% to 7.34%, daily AI use among US employees rose from about 4% to 15%, and PwC’s advertised premium for AI skills rose from 56% to 62%. Employers were cutting some roles in AI’s name and paying more for AI skills in others at the same time.

The measures that diverge are the most informative. Software-development postings were flat at about a third below their 2020 level through 2024 and 2025, then recovered to 18.4% below in October 2026, while the Stanford early-career gap in AI-exposed jobs widened from 16% to 19%. AI-attributed cuts peaked in May 2026 and fell to 1,265 in October. And the share of executives reporting any employment effect from AI remains under 10%. The comparison does not support a story of steady, economy-wide displacement; it supports one of reallocation towards experienced workers and AI-skilled roles. Productivity, the indicator that would most justify further cuts, has moved least: self-reported time savings rose from about 1.4% to about 2.2% of US work hours between November 2024 and the first quarter of 2026.

Exhibit 10
Every AI-at-work indicator rose between 2024 and 2026 except software postings, which fell and then recovered, and the AI layoff count, which peaked in May 2026
Key AI-and-work indicators by calendar year described; 2026 values are latest to date
Indicator202420252026 (to date)
AI-attributed US job cuts (Challenger)12,74254,836121,401 (Jan–Oct)
AI share of all announced cuts1.7%4.5%19.7%
US Indeed postings mentioning AI2.71% (Dec)4.44% (Dec)7.34% (Sep)
US software-development postings vs Feb 2020−32.5% (Dec)−32.7% (Dec)−18.4% (Oct)
US employees using AI daily (Gallup)~4% (Q2)12% (Q4)15% (May)
AI skills wage premium in job ads (PwC)56%62%—
Early-career gap, AI-exposed jobs (Stanford)—16%19%
Work hours saved by generative AI (St. Louis Fed)1.4%~2.0%~2.2% (Q1)
Hiring managers who rehired an AI-cut role——32%
Source: Challenger, Gray & Christmas; Indeed Hiring Lab; Gallup; PwC; Stanford Digital Economy Lab; St. Louis Fed; Robert Half; Report AI analysis MEDIUM
Notes: Each row is one series; compare across columns, not across rows. Share-of-cuts row is DERIVED. Gallup changed its question in Q3 2025. PwC values are by year of ads analysed (2025 = 2026 edition); PwC restated 2024 as 57%. Stanford 2025 value is the Nov 2025 revision (RESTATED from 13%). St. Louis Fed 2025–26 values read from a secondary summary.REPORTAI report-ai.org

Year over year, every comparable measure

Measure202420252026 (to date)2025→26
US job cuts attributed to AI (Challenger)12,74254,836121,401 (Jan–Oct, to date)+121%
All announced US job cuts (Challenger)761,3581,206,374617,189 (Jan–Oct, to date)-49%
AI-attributed cuts as a share of all US announced cuts1.7%4.5%19.7% (Jan–Oct, to date)+338%
Cumulative US job cuts attributed to AI since 2023 (Challenger)16,98971,825193,226 (to date)+169%
Share of US job postings on Indeed mentioning AI-related terms2.71%4.44%7.34% (to date)+65%
Share of UK job postings on Indeed mentioning AI-related terms3.44%7.12%9.92% (to date)+39%
US software-development job postings on Indeed vs 1 Feb 202067.5 (−32.5%)67.3 (−32.7%)81.6 (−18.4%, to date)+21%
All US job postings on Indeed vs 1 Feb 2020110.6103.4103.8 (to date)+0%
US employees using AI at work daily (Gallup)4%12%15% (to date)+25%
Share of all US work hours saved by generative AI (St. Louis Fed survey)1.4%about 2.0%about 2.2% (to date)+8%
  • AI-attributed US job cuts rose from 12,742 in 2024 to 54,836 in 2025 and 121,401 in January–October 2026; as a share of all announced cuts, from about 1.7% to 4.5% to 19.7% (Report AI’s division of Challenger figures).
  • The share of US Indeed postings mentioning AI rose from 2.71% at the end of 2024 to 4.44% at the end of 2025 and 7.34% on 30 September 2026, about 2.7 times in 21 months.
  • US software-development postings sat about 32.5% below their February 2020 level at the end of both 2024 and 2025, then recovered to 18.4% below by 30 October 2026: AI did not stop developer hiring, but the recovery favours senior roles while early-career employment in exposed occupations kept falling behind.
  • PwC’s AI skills wage premium in job advertisements rose from 25% (2023 ads) to 56% (2024 ads, restated 57%) to 62% (2025 ads), and daily AI use among US employees rose from about 4% (Q2 2024) to 12% (late 2025) to 15% (May 2026).

All eleven themes side by side: State of AI, year over year.

Data appendix

The numbers, 2024 to 2026

Every measure in this report with its value for each year, as-of date, source and confidence. Change compares 2025 with 2026 to date where both are numeric.

AI-attributed layoffs

Measure202420252026 (to date)Change
US job cuts attributed to AI (Challenger)
announced job cuts
12,742
Dec 2024 (full year), Challenger, Gray & Christmas (via secondary reporting) MEDIUM
54,836
Dec 2025 (full year), Challenger, Gray & Christmas, 2025 year-end report MEDIUM
121,401 (Jan–Oct, to date)
5 Nov 2026 (Jan–Oct 2026), Challenger, Gray & Christmas, October 2026 report MEDIUM
+121%
All announced US job cuts (Challenger)
announced job cuts
761,358
Dec 2024 (full year), Challenger, Gray & Christmas (2025 year-end report, prior-year comparison) MEDIUM
1,206,374
Dec 2025 (full year), Challenger, Gray & Christmas, 2025 year-end report MEDIUM
617,189 (Jan–Oct, to date)
5 Nov 2026 (Jan–Oct 2026), Challenger, Gray & Christmas, October 2026 report MEDIUM
-49%
AI-attributed cuts as a share of all US announced cuts
% of announced cuts
1.7%
Dec 2024 (full year), Report AI, from Challenger figures DERIVED
4.5%
Dec 2025 (full year), Report AI, from Challenger figures (Challenger also states 4.5%) DERIVED
19.7% (Jan–Oct, to date)
5 Nov 2026 (Jan–Oct 2026), Report AI, from Challenger figures DERIVED
+338%
Cumulative US job cuts attributed to AI since 2023 (Challenger)
announced job cuts (cumulative)
16,989
Dec 2024, Report AI, from Challenger figures DERIVED
71,825
Dec 2025, Challenger, Gray & Christmas, 2025 year-end report (via secondary reporting) MEDIUM
193,226 (to date)
5 Nov 2026 (through Oct 2026), Report AI, from Challenger figures DERIVED
+169%

Reversals and rehiring

Measure202420252026 (to date)Change
US hiring managers who cut a role because of AI and later rehired for it
% of hiring managers
——32%
2026, Robert Half survey (~2,000 US hiring managers) MEDIUM
Employers who regret AI-related layoffs
% of employers
——55%
2026, Forrester, Future of Work / Predictions 2026 MEDIUM

Entry-level and graduate hiring

Measure202420252026 (to date)Change
Relative employment decline, ages 22–25, most AI-exposed US occupations (Stanford DEL)
% relative to less-exposed peers
—16%
Nov 2025 (ADP payroll data, late 2022 to 2025), Brynjolfsson, Chandar & Chen, ‘Canaries in the Coal Mine?’, Stanford Digital Economy Lab, revised 13 Nov 2025 MEDIUM
19%
12 Aug 2026 (payroll data through Jun 2026), Stanford Digital Economy Lab, Canaries August 2026 update MEDIUM
+19%
Change in entry-level software developer job postings since 2022
% change
——−67%
2026 (2022–2026), Industry posting analyses compiled by Report AI MEDIUM
New-graduate hiring at 12 large US tech firms vs 2019 (SignalFire)
% change vs 2019
about −50%
2024 (SignalFire State of Talent 2025), SignalFire, State of Tech Talent Report 2025 MEDIUM
—about −65%
Jun 2026, SignalFire, State of Tech Talent Report 2026 / June 2026 update MEDIUM
US employers’ projected change in hiring of new college graduates (NACE)
% change vs prior class (employer projection)
—+0.6% (class of 2025)
Spring 2025 update, NACE, Job Outlook 2025 Spring Update MEDIUM
+5.6% (class of 2026)
Apr 2026 (fielded 12 Feb–17 Mar 2026), NACE, Job Outlook 2026 Spring Update MEDIUM
+833%

AI job postings

Measure202420252026 (to date)Change
Share of US job postings on Indeed mentioning AI-related terms
% of postings
2.71%
31 Dec 2024, Indeed Hiring Lab, AI Tracker data (7-day trailing average) HIGH
4.44%
31 Dec 2025, Indeed Hiring Lab, AI Tracker data HIGH
7.34% (to date)
30 Sep 2026 (latest in dataset), Indeed Hiring Lab, AI Tracker data HIGH
+65%
Share of UK job postings on Indeed mentioning AI-related terms
% of postings
3.44%
31 Dec 2024, Indeed Hiring Lab, AI Tracker data HIGH
7.12%
31 Dec 2025, Indeed Hiring Lab, AI Tracker data HIGH
9.92% (to date)
30 Sep 2026, Indeed Hiring Lab, AI Tracker data HIGH
+39%

Software developers

Measure202420252026 (to date)Change
US software-development job postings on Indeed vs 1 Feb 2020
index (1 Feb 2020 = 100), seasonally adjusted
67.5 (−32.5%)
31 Dec 2024, Indeed Hiring Lab, Job Postings Index by sector HIGH
67.3 (−32.7%)
31 Dec 2025, Indeed Hiring Lab, Job Postings Index by sector HIGH
81.6 (−18.4%, to date)
30 Oct 2026, Indeed Hiring Lab, Job Postings Index by sector HIGH
+21%
All US job postings on Indeed vs 1 Feb 2020
index (1 Feb 2020 = 100), seasonally adjusted
110.6
31 Dec 2024, Indeed Hiring Lab, Job Postings Index HIGH
103.4
31 Dec 2025, Indeed Hiring Lab, Job Postings Index HIGH
103.8 (to date)
30 Oct 2026, Indeed Hiring Lab, Job Postings Index HIGH
+0%
Developer hours per week reviewing AI-generated code vs writing new code
hours per week
——11.4h reviewing vs 9.8h writing
Q1 2026, Developer survey data compiled by Report AI MEDIUM

Worker AI use

Measure202420252026 (to date)Change
US employees using AI at work daily (Gallup)
% of employees
4%
Q2 2024, Gallup Workforce survey MEDIUM
12%
Fall 2025 (Q4), Gallup Workforce survey (AP reporting) MEDIUM
15% (to date)
May 2026 (Q2), Gallup, AI indicator MEDIUM
+25%
US employees using AI at work at least a few times a year (Gallup)
% of employees
—46%
Q4 2025, Gallup Workforce survey MEDIUM
52% (to date)
May 2026 (Q2), Gallup, AI indicator MEDIUM
+13%
US workers saying at least some of their work is done with AI (Pew)
% of workers
16%
Oct 2024, Pew Research Center MEDIUM
21%
Sep 2025, Pew Research Center MEDIUM
—

AI skills premium

Measure202420252026 (to date)Change
Wage premium in job ads requiring AI skills vs comparable roles (PwC Global AI Jobs Barometer)
% premium
56%
2024 job ads (PwC 2025 Barometer, Jun 2025), PwC, 2025 Global AI Jobs Barometer MEDIUM
62%
2025 job ads (PwC 2026 Barometer, published 15 Jun 2026), PwC, 2026 Global AI Jobs Barometer MEDIUM
—

Productivity evidence

Measure202420252026 (to date)Change
Share of all US work hours saved by generative AI (St. Louis Fed survey)
% of total work hours
1.4%
Nov 2024 survey, Bick, Blandin & Deming / Federal Reserve Bank of St. Louis, Real-Time Population Survey MEDIUM
about 2.0%
Q4 2025, FRED series RPSGENAITSALL (St. Louis Fed RPS), via secondary summary MEDIUM
about 2.2% (to date)
Q1 2026, FRED series RPSGENAITSALL, via secondary summary MEDIUM
+8%
Task-time effect of early-2025 AI tools on experienced open-source developers (METR RCT)
% change in completion time
—+19% (slower)
Feb–Jun 2025 trial, published 10 Jul 2025, METR, ‘Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity’ MEDIUM
—

Customer support

Measure202420252026 (to date)Change
Median tier-1 customer-support deflection by AI, enterprise programmes
% of tier-1 contacts resolved without a human
——41.2%
2026, ClarityArc 2026 production benchmarks; Zendesk CX Trends 2026 aggregate MEDIUM

Occupational exposure

Measure202420252026 (to date)Change
Share of global employment exposed to AI (IMF)
% of employment
40%
Jan 2024, IMF, ‘Gen-AI: Artificial Intelligence and the Future of Work’ MEDIUM
——
Task-automation share, office and administrative support (highest of any occupational group)
% of tasks
——46%
2026, Studies compiled on Report AI’s by-occupation page MEDIUM

Forecasts

Measure202420252026 (to date)Change
Jobs created worldwide by 2030 (WEF forecast)
jobs (forecast, 2025–2030)
—170M
Jan 2025, World Economic Forum, Future of Jobs Report 2025 MEDIUM
170M (unchanged)
Nov 2026 (no newer edition), World Economic Forum, Future of Jobs Report 2025 MEDIUM
+0%
Jobs displaced worldwide by 2030 (WEF forecast)
jobs (forecast, 2025–2030)
—92M
Jan 2025, World Economic Forum, Future of Jobs Report 2025 MEDIUM
92M (unchanged)
Nov 2026 (no newer edition), World Economic Forum, Future of Jobs Report 2025 MEDIUM
+0%
Net change in jobs worldwide by 2030 (WEF forecast)
jobs (forecast, 2025–2030)
—+78M
Jan 2025, World Economic Forum, Future of Jobs Report 2025 MEDIUM
+78M (unchanged)
Nov 2026 (no newer edition), World Economic Forum, Future of Jobs Report 2025 MEDIUM
+0%
Fact check

Numbers to stop quoting

Widely repeatedWhat the evidence supports
“AI is the top reason for US layoffs in 2026.”AI led Challenger’s stated reasons for several months from March 2026, but in Challenger’s 5 November 2026 report ‘market/economic conditions’ (130,531 cuts) overtook AI (121,401) for January–October, and AI-attributed cuts fell to 1,265 in October. · detail
“AI will wipe out 92 million jobs by 2030.”The World Economic Forum’s Future of Jobs Report 2025 forecasts 92 million jobs displaced and 170 million created worldwide by 2030, a net gain of 78 million, from all structural trends (technology, green transition, demographics, economics), not AI alone. It is an employer-survey forecast, not a measurement. · detail
“Workers with AI skills earn a 56% wage premium (or ‘up to 25%’).”Superseded. PwC’s 2026 Global AI Jobs Barometer (June 2026) puts the average premium in job advertisements at 62%; the 56% describes 2024 ads and was restated as 57%; 25% describes 2023 ads. It is an advertised-wage gap between comparable roles, not a measured pay rise for individual workers. · detail
“Stanford found entry-level jobs in AI-exposed occupations are down 13%.”Superseded. The August 2025 working paper’s 13% was revised to 16% in November 2025, and the August 2026 update puts employment of 22-to-25-year-olds in AI-exposed occupations 19% below the path of less-exposed peers. It is a relative, descriptive measure, not a count of jobs lost to AI. · detail
“Half of AI layoffs have already been reversed.”That is Forrester’s forecast for the end of 2026, which resolves in January 2027. The interim evidence is that 32% of US hiring managers who cut a role because of AI have rehired for it (Robert Half, 2026). · detail
“Software developer job postings are still collapsing because of AI.”Indeed’s US software-development postings index was flat at about 67 (one-third below February 2020) through 2024 and 2025, then rose to 81.6 on 30 October 2026, up 21% in ten months. The steeper ‘entry-level postings −67% since 2022’ figure is a vendor estimate for junior roles only. · detail
Outlook

Whether 2026’s fall in AI-attributed cuts is a pause or a turn will be visible in the first half of 2027

The indicators to watch in 2027 are the ones that would change the reading in this report. If AI-attributed cuts return above 10,000 a month, the October 2026 low was a pause in restructuring; if they stay below about 5,000 a month through the first quarter, the 2026 peak looks like a one-off wave. If the Stanford early-career gap widens beyond about 25%, the entry-level squeeze is deepening rather than stabilising. Forrester’s reversal forecast resolves in January 2027, and the 2027 PwC Barometer will show whether the advertised AI-skill premium is still rising or beginning to compress as AI skills become common.

Two longer-run signals matter more than any single month. The first is whether firms begin to report measurable productivity effects: in early 2026, 89% of executives in NBER’s survey reported none. The second is whether the recovery in software postings reaches junior roles; so far it has been concentrated among senior engineers, which leaves the pipeline problem described in Chapter 3 unresolved. A new WEF Future of Jobs edition, if published, will show whether employers still expect net job creation by 2030. Each signal below is stated with its current value and date, so that the next edition of this report can say plainly which way it moved and whether our reading should change.

What to watch in 2027

IndicatorWhere it standsWhat would change the picture
Challenger AI-attributed US job cuts, monthly1,265 in October 2026; 121,401 January–October 2026 (5 Nov 2026)Back above 10,000 a month would mean restructuring resumed; below 5,000 a month through March 2027 would mark 2026 as a peak
AI’s rank among Challenger’s stated reasons, year to dateSecond to ‘market/economic conditions’ (121,401 vs 130,531, Jan–Oct 2026)AI regaining first place in 2027 would indicate cuts are again being framed as AI-driven
Stanford early-career employment gap, ages 22–25, AI-exposed jobs19% (August 2026 update, data through June 2026)Above 25% would indicate a deepening squeeze; below 15% would suggest stabilisation
Indeed US software-development postings index (Feb 2020 = 100)81.6 on 30 October 2026Above 90 with junior roles sharing the gain would ease the pipeline concern; a fall back below 70 would reverse the 2026 recovery
Share of US Indeed postings mentioning AI7.34% on 30 September 2026Above 10% would put US AI demand at the UK’s current level; a flattening would suggest saturation
Rehiring after AI cuts (Forrester forecast)32% of hiring managers have rehired an AI-cut role (Robert Half, 2026); Forrester forecasts ~50% reversed by end-2026Credible data showing ~50% refilled or reversed would confirm the forecast when it resolves in January 2027
PwC AI skills wage premium (2027 edition)62% (2026 edition, published June 2026)A fall below 50% would suggest AI skills are becoming commoditised
Executives reporting any productivity effect from AI (NBER)89% reported none over the previous three years (February 2026)A majority reporting measurable effects would mark the start of aggregate gains
Definitions
AI-attributed job cut
A job cut announced by a US employer for which the employer named artificial intelligence as the reason, as classified by Challenger, Gray & Christmas. It counts planned cuts at announcement, not completed separations, and depends on the reason the employer chooses to give.
Exposure
An estimate of how much of an occupation’s work current AI could perform or assist. It is not a forecast of job loss; exposed jobs may be augmented rather than replaced.
Task-automation share
The share of an occupation’s tasks that a study judges automatable. A 46% task share does not mean 46% of jobs are lost.
Relative employment decline
In the Stanford Canaries studies, how far employment of a group (here, workers aged 22–25 in the most AI-exposed occupations) has fallen behind a comparison group since late 2022. It is descriptive, not a causal estimate.
Job postings index
Indeed Hiring Lab’s seasonally adjusted measure of job postings on Indeed relative to 1 February 2020 (= 100). It measures advertised demand, not hires.
AI skills wage premium
In PwC’s Global AI Jobs Barometer, the average difference in advertised wages between job ads that require AI skills and comparable ads that do not. It is not a measured pay rise for individual workers.
Deflection
The share of customer contacts an AI system resolves without passing them to a human. Vendors measure it differently, which is why published figures range from about 20% to 90%.
Reversal
Any rehiring, re-scoping or cancellation of an AI-justified job cut. Definitions vary across surveys, which makes forecasts such as Forrester’s hard to score.

Frequently asked

How many US jobs have been cut because of AI?

Challenger, Gray & Christmas counted 12,742 US job cuts attributed to AI in 2024, 54,836 in 2025 and 121,401 from January to October 2026, about 193,000 since its tracking began in 2023 by Report AI’s sum. These are announced cuts where the employer named AI as the reason; they can overstate AI’s role, because AI is a palatable public reason, and understate it, because some cuts are filed under other headings. AI-attributed cuts fell to 1,265 in October 2026, and by 5 November 2026 market and economic conditions had overtaken AI as the leading stated reason for 2026.

Are companies rehiring workers after AI layoffs?

Yes, in measurable numbers. Robert Half found in 2026 that 32% of US hiring managers who eliminated a role primarily because of AI later rehired for it, usually in a hybrid role paying 20–35% more. Klarna began rehiring human support agents in 2025 after its CEO said cost had been too predominant a factor, and Meta cancelled the second wave of its 2026 Project OT cuts. Forrester has forecast that about half of AI-attributed layoffs will be reversed in some form by the end of 2026; that forecast resolves in January 2027.

Is AI reducing entry-level and graduate jobs?

The evidence points that way for AI-exposed occupations, though causation is not proven. Stanford Digital Economy Lab’s August 2026 update, using ADP payroll data through June 2026, found employment of 22-to-25-year-olds in AI-exposed US occupations 19% below where it would have been had it kept pace with less-exposed peers, up from 16% in November 2025. SignalFire reports new-graduate hiring at 12 large tech firms about 65% below 2019 in 2026. Graduate hiring overall is not collapsing: NACE’s spring 2026 update projected a 5.6% increase in hiring for the class of 2026.

How many workers use AI at work, and does it pay?

Gallup found in May 2026 that 52% of US employees used AI at work at least a few times a year, 30% a few times a week or more and 15% daily, up from about 4% daily in mid-2024. PwC’s 2026 Global AI Jobs Barometer, published in June 2026 and based on the previous year’s job advertisements across many countries, found that jobs requiring AI skills advertised wages 62% higher on average than comparable roles without them, up from 56% (restated 57%) a year earlier. The premium varies widely by sector, from 118% in consumer markets to 16% in government.

Does AI actually make workers more productive?

On specific tasks often, across whole jobs not yet measurably. The St. Louis Fed’s survey put time saved by generative AI at about 1.4% of all US work hours in November 2024 and about 2% by early 2026, self-reported. A METR randomised trial published in July 2025 found experienced developers took 19% longer with early-2025 AI tools on codebases they knew. In NBER’s February 2026 study of executives in four countries, 89% reported no effect of AI on labour productivity over the previous three years, while expecting a 1.4% gain over the next three.

Methodology & sources

Figures are compiled from company disclosures, regulators, research firms and reputable reporting, and cross-checked against Report AI’s own index pages. Each value carries its as-of date, source and a confidence rating: HIGH where the primary source was read, MEDIUM where it rests on consistent reputable secondary reporting (most figures, because many primary pages cannot be opened from our research environment), DERIVED where it is our own arithmetic, and CONTESTED where credible sources disagree. A year means the calendar year the figure describes; 2026 values are the latest available and marked “to date”. Weekly and monthly user counts, and forecasts and outcomes, are never compared as if they were the same measure. Exhibits plot only dated, sourced readings; lines are not interpolated between them, and values with different definitions are never placed on one axis. Known gaps in this edition: Primary sources could not be opened: WebFetch failed (stackoverflow.blog not resolvable), FRED and Challenger’s site were blocked; only Indeed Hiring Lab’s GitHub datasets were read directly (HIGH). The session’s WebSearch budget ran out before a few checks (METR follow-up study, exact FRED RPSGENAITSALL values, Stack Overflow 2026 headline figures).; Challenger 2024 AI-attributed total (12,742) comes from a single secondary summary; it is consistent with Challenger’s cumulative 71,825 since 2023 only if 2023 was 4,247, which was not independently confirmed.; Gallup ‘any use’ share for 2024 not found; the 2024 daily-use value (4%) is taken from Gallup’s statement that daily use doubled from 4% to 8% in the year to Q2 2025. Gallup changed its question (added ‘don’t know’) in Q3 2025.; Stack Overflow 2026 Developer Survey (published Oct 2026): no all-respondent AI-use or trust figure retrieved. 2024→2025 trust figures are themselves inconsistent across coverage (43%→33% vs 40%→29%), so developer AI adoption was left out of the metrics.; PwC year mapping: the 2025 edition explicitly analyses 2024 ads; that the 2026 edition (62%) analyses 2025 ads is assumed. No 2026-year value until the 2027 edition.; No newer WEF Future of Jobs edition after January 2025 was found; WEF 2030 values are repeated for 2026 as the forecast in force, not as new data.; Pew’s 2026 update and NACE’s class-of-2024 projection were not found. NACE’s class-of-2026 figures mix mean (+1.6%, Nov 2025) and median (+5.6%, Apr 2026) bases.; Lightcast/Stanford AI Index US AI-postings share for 2025 varies by write-up (2.5%, 2.56%, PwC’s 2.8%); left as a note on the Indeed metric rather than a separate series.; No 2026 randomised productivity study with measured whole-job effects was found; the 2023 customer-support field study (14% more issues resolved per hour, data 2020–21) predates the series window.; The 55% ’employers regret AI layoffs’ figure is credited to Forrester on the site; secondary reporting links it also to an Orgvue survey — origin should be confirmed.; Stale figures on the site (corpus 10 Nov 2026): pages 387 and 395 show the PwC premium as 56% (current: 62%); page 395 shows Stanford −13% (revised to 16%, now 19%); page 1114 says ‘up to about 25%’; page 88 (updated July 2026) uses H1 2026 Challenger figures (~101,743) and the 16% Stanford figure; pages 2297 and 2300 present AI as the #1 layoff reason, true for March–mid 2026 but not year to date as of 5 Nov 2026.; Indeed’s AI tracker values are revised over time: Hiring Lab reported 4.2% for end-2025 in January 2026; the current dataset shows 4.44% (marked RESTATED).; The ‘58% of employees use AI semi-regularly’ figure on page 387 (Stanford HAI 2026, global) was not traced to its underlying survey and is not used here.; Occupational exposure figures (page 2300) come from studies with different methods and were gathered via search summaries; only the office-and-admin task share is carried as a metric.. These reports compile and verify secondary data with original analysis; they are not original surveys. See our editorial standards, corrections log and fact-check register.