The Labour Market Impact of AI

This assessment covers the period from 30 November 2022 to 2 September 2026. Everything below is dated, and the distinction between what has been measured and what has merely been projected is the point of the exercise.

TL;DR

The measured net effect of AI on jobs to date is small but sharply uneven: there is no economy-wide displacement wave, but entry-level white-collar hiring is visibly contracting (Stanford finds the employment gap for 22 to 25 year olds in the most AI-exposed jobs widened to 19% by August 2026, and Goldman Sachs estimates AI is subtracting a net 16,000 US jobs a month), while AI-skilled workers command a 56% to 62% wage premium and demand for AI roles keeps rising.
The big job-creation numbers are almost all projections, not measurements: the WEF’s “net 78 million new jobs by 2030” and the data-centre “4.7 million construction jobs” figures are forecasts from advocacy-linked or survey models, whereas the hard measured numbers (AI-cited layoffs, AI job-posting shares, graduate unemployment) are far smaller and more concrete.
For the three groups this affects most, the verdict is: new graduates face the worst entry-level market in over a decade and must show applied AI skill to stand out; current students should keep studying computing but pivot to AI-augmented, judgement-heavy specialisations; and existing knowledge workers are being augmented far more than replaced, with the AI-fluent pulling away on pay.

Key Findings

Job creation is dominated by forecasts. The single most-cited figure, the World Economic Forum’s Future of Jobs Report 2025 projection of 170 million jobs created and 92 million displaced for a net 78 million (a 7% net rise) by 2030, is a survey-based forecast, not a measurement, drawn from over 1,000 employers across 22 industries and 55 economies representing more than 14 million workers.
AI-cited layoffs remain a small share of total layoffs. Challenger, Gray and Christmas, the only tracker that isolates AI as a stated reason, counted 54,836 AI-attributed US layoffs in 2025 (about 5% of all cuts) and 71,825 cumulatively since it began tracking in 2023.
Entry-level is where the damage concentrates. Stanford’s “Canaries in the Coal Mine” and Goldman Sachs both find the harm falls on the young, while experienced workers are stable or growing.
CEO rhetoric has run ahead of company hiring. Several firms that loudly announced AI-driven cuts (Klarna, IBM, Salesforce) have since rehired humans or grown headcount; the words and the actions frequently diverge.
AI is augmenting more than automating, for now. Real usage data from Anthropic and Microsoft shows AI concentrated in information work (coding, writing, information gathering) and used collaboratively about as often as autonomously.
The wage premium is the clearest positive signal. AI-skilled workers earn a large and growing premium, and AI-exposed firms are growing headcount faster than less-exposed ones.

Details

1. Jobs created (direct and indirect): almost entirely projected, not measured

Global projection (WEF). The Future of Jobs Report 2025 (published 8 January 2025) projects 170 million new jobs created and 92 million displaced by 2030, a net increase of 78 million or about 7% of employment. Fastest-growing roles include big-data specialists, fintech engineers, and AI and machine-learning specialists. WEF also states investment in generative AI has increased eightfold since ChatGPT’s launch, and that 39% of core skills will change by 2030. This is a forecast from employer surveys, not measured job creation.

Direct AI roles (measured, job-posting based). The Stanford AI Index (produced with Lightcast) found AI-related job postings rose from 1.4% of all US postings in 2023 to 1.8% in 2024. The 2026 edition reports AI skills now appear in 2.5% of all US postings, up 55% year on year, and that mentions of the “agentic AI” skill cluster rose over 280% in a year, from 0.06% to 0.23% of US postings, roughly 90,000 postings. Generative-AI skill mentions rose from about 16,000 postings in 2023 to more than 66,000 in 2024. PwC’s Global AI Jobs Barometer found the number of AI jobs almost twice as high in its 2026 edition as in 2024, with AI-skill roles growing about 9% (roughly eight times the wider market’s rate).

Indirect: data centres, chips, energy (mostly projected, and contested). The most-repeated figure, 4.7 million temporary US construction jobs plus roughly 697,000 permanent operations jobs, comes from a 2025 report by the American Edge Project, a policy advocacy group formed by Meta; this is a single-origin, industry-advocacy projection and should be treated with caution. A separate projection from Kevin McCarthy’s ALFA Institute (via Capital Policy Analytics) estimates a $100 billion data-centre investment would create 500,000 jobs, mostly in manufacturing. Measured evidence is far more modest: Brookings economists Dany Bahar and Greg Wright found counties receiving their first large data centre saw private employment rise 4% to 5% over five to six years, but Wright concluded in November 2025 that the net county employment effect is “effectively zero” because workers shift between subsectors. Virginia’s data-centre industry added just 1,610 jobs in fiscal 2025 against $1.9 billion in tax benefits ($1.2 million per job). Chip fabs are more labour-intensive: TSMC’s Arizona project is projected to create about 6,000 direct manufacturing jobs and over 20,000 construction jobs.

India (measured and projected). Nasscom projects India will add over 3.5 lakh (350,000) technology jobs in 2025 to 2026, with freshers about 45% of new hires. Nasscom has also projected the cloud ecosystem could generate 1.4 crore (14 million) jobs and that cloud could reach 8% of GDP by 2026. India’s tech workforce is expected to reach 7.5 million by 2030.

Verdict: The honest answer is that no one has a credible measured total of jobs “created by AI” since November 2022. The defensible measured signals are the rise in AI job-posting shares (from about 1.4% to 2.5% of US postings) and the near-doubling of AI-specific roles. Everything at the “millions of jobs” scale is a projection, and the data-centre construction numbers rest heavily on one industry-advocacy source.

2. Layoffs where AI was cited

The authoritative tracker is Challenger, Gray and Christmas, which began recording AI as a stated reason in 2023.

Period	AI-cited job cuts (Challenger, US)	Context
2023 (first tracked) to end-2024	roughly 17,000 (implied residual)	AI a minor stated reason
2025 full year	54,836 (about 5% of all 2025 cuts)	Total 2025 cuts: 1,206,374, up 58% on 2024’s 761,358
Cumulative since 2023 (per 2025 Year-End Report, 8 January 2026)	71,825
2026 to date (through March)	12,304 (8% of cuts; AI ranked as a leading stated reason)

Andy Challenger’s own caution matters: “Naming AI in a layoff announcement can win over investors while pushing current and prospective employees away. That’s why the messaging has swung from hedging to aggressively citing it.” Challenger also uses a separate category, “Technological Update (possibly AI),” which accounted for 20,219 cuts in 2025, suggesting the true AI-linked figure may be higher but is deliberately obscured by companies. These are announced US cuts only, a single-origin national dataset.

Largest individual AI-linked layoffs (explicitly cited or strongly implied):

IBM: about 8,000 roles from 2023 onward, concentrated in HR, as its AskHR AI platform automated tasks; about 200 HR jobs were taken over by AI agents. Crucially, CEO Arvind Krishna says total employment rose because savings were reinvested in software, sales and marketing.
Salesforce: cut about 4,000 customer-service roles; Benioff said “I need less heads” (September 2025).
Amazon: more than 27,000 corporate roles cut since 2022; Jassy’s June 2025 memo tied future reductions to AI.
Klarna: AI assistant said to do the work of 700 agents; workforce shrank from about 5,500 to about 3,000, later partly reversed.
CrowdStrike: 5% of workforce (May 2025), citing AI “efficiencies.”
Snap: about 1,000 staff (16%), April 2025.
Workday: more than 1,700 in 2025 while prioritising AI.
Duolingo, Dropbox, Chegg, Intuit, UPS: cited AI or restructuring in cuts; Meta cut Reality Labs staff while pivoting to AI.

Verdict: Measured AI-attributed layoffs are real but modest, about 71,825 cumulatively in the US, dwarfed by macro and restructuring cuts. AI is a growing but still minority stated cause, and companies both over-claim (for investors) and under-claim (to avoid backlash), so the true figure is genuinely uncertain.

3. Tech CEO predictions of AI-led job replacement

Dario Amodei (Anthropic), May 2025, Axios: AI could “wipe out half of all entry-level white-collar jobs” and spike unemployment to “10 to 20% in the next one to five years.” “We, as the producers of this technology, have a duty and an obligation to be honest about what is coming.” By May 2026 he had softened, invoking the Jevons Paradox at a JPMorgan event; and in July 2026 Anthropic’s own head of economics, Peter McCrory, published an analysis arguing AI has caused no material rise in US unemployment, a direct in-house rebuttal.
Andy Jassy (Amazon), 17 June 2025 memo: “We will need fewer people doing some of the jobs that are being done today, and more people doing other types of jobs… in the next few years, we expect that this will reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company.”
Marc Benioff (Salesforce): February 2025, “We’re not going to hire any new engineers this year” (citing a 30% engineering productivity gain) and “we are the last generation to manage only humans.” September 2025, “I need less heads” on cutting 4,000 support roles. Then April 2026, reversed tone: “We’re hiring 1,000 new grads and interns right now to ride the AI exponential.”
Mark Zuckerberg (Meta): said AI could “effectively be a sort of mid-level engineer” able to write code in 2025.
Sundar Pichai (Google): said in October 2024 that AI writes more than 25% of new code at Google.
Tobi Lütke (Shopify), April 2025 memo: teams must “demonstrate why they cannot get what they want done using AI” before requesting more headcount.
Luis von Ahn (Duolingo): “We’ll gradually stop using (human) contractors to do work that AI can handle,” declaring Duolingo “AI-first” (April 2025), later walked back amid backlash.
Micha Kaufman (Fiverr), May 2025 memo: “AI is coming for your jobs. Heck, it’s coming for my job too.”
Sam Altman (OpenAI): has argued for “realistic optimism” but concedes displacement will come; in early 2026 warned “AI washing” (blaming AI for unrelated cuts) is real, while maintaining measurable displacement is only a matter of time.
Jamie Dimon (JPMorgan): appeared alongside Amodei; separately emphasised job creation such as 300,000 shipbuilding workers.

Verdict: The loudest 2025 predictions (Amodei, Benioff) have been partly walked back or contradicted by the same firms’ hiring in 2026. The pattern is prediction inflation followed by quiet moderation once data failed to show mass displacement.

4. Companies that said AI will cut jobs versus what they then did

Company	Date of statement	What they said	Subsequent hiring or headcount	Words matched actions?
Klarna	2023 to Feb 2024	AI does the work of 700 agents; CEO said “AI can already do all the jobs we humans do”	By mid-2025 rehiring human agents; CEO Siemiatkowski: “we went too far”, cited “lower quality”; workforce still cut from about 5,500 to about 3,000	No, notable public reversal
IBM	2023 to 2025	Replaced about 8,000 (mainly HR) with AskHR AI; 200 HR roles to AI agents	Total employment “actually gone up”; reinvested in software, sales, marketing	Partly, cut and grew simultaneously
Salesforce	Feb 2025	“Not going to hire any new engineers this year”; cut 4,000 support roles	April 2026 hiring 1,000 new grads and interns; adding thousands of sales staff	No, reversed within about a year
Duolingo	April 2025	“AI-first”, phasing out human contractors	Walked back messaging after public backlash	Partly, softened
Shopify	April 2025	Prove AI cannot do the job before hiring humans	Policy retained; no mass rehire reported	Yes, sustained
Amazon	June 2025	Corporate workforce to shrink due to AI	Continued corporate cuts; still hiring in AI and operations	Broadly yes so far
Microsoft	2025	Cut roles while investing heavily in AI	Ongoing AI hiring; large layoffs in 2025	Mixed
Meta	Feb 2025, 2026	AI as mid-level engineer; Reality Labs cuts	Aggressive AI-talent hiring at high pay	Mixed, cutting some, hiring AI

Count: At least eight named large companies publicly stated AI would reduce jobs. Of these, at least three (Klarna, IBM, Salesforce) demonstrably rehired humans or grew headcount after the statement, and Duolingo softened its stance. The clearest single-source reversal narrative is Klarna’s, repeated across many outlets from one Bloomberg interview.

The reversals are part of a measured pattern, not isolated missteps. Gartner (press release dated 3 February 2026, per Senior Director Analyst Kathy Ross) predicts: “By 2027, 50% of companies that attributed headcount reduction to AI will rehire staff to perform similar functions, but under different job titles”; a Gartner survey of 321 customer-service leaders (October 2025) found only 20% had actually reduced agent staffing due to AI. And the IBM 2025 CEO Study (2,000 CEOs across 24 industries and 33 countries, surveyed Q1 2025) found only 25% of AI initiatives had delivered the expected ROI over the last few years, and only 16% had scaled enterprise-wide.

Verdict: Actions have lagged words. The high-profile reversals (Klarna especially) show that full automation of relational, high-stakes work has repeatedly failed on quality, forcing rehiring.

5. Which jobs are most and least impacted, and which are hiring

Exposure and applicability rankings (theoretical, from usage data):

Microsoft Research (Tomlinson et al., 2025, 200,000 Bing Copilot conversations mapped to O*NET): highest AI applicability for translators, historians, writers, customer-service and sales representatives, and other information-work roles (creating, processing and communicating information). Microsoft stressed high applicability does not equal replacement.
Anthropic Economic Index: computer and mathematical tasks dominate real Claude usage, about a third of Claude.ai conversations and nearly half of API traffic.

Measured employment-change rankings:

Stanford “Canaries in the Coal Mine” (Brynjolfsson, Chandar, Chen; August 2025, revised February and August 2026, using ADP payroll data on 4.6 million workers across 730-plus occupations): workers aged 22 to 25 in the most AI-exposed occupations saw a 13% relative employment decline (first paper); by the August 2026 update the young-worker employment gap had widened to 19%. Software developers aged 22 to 25 fell nearly 20% from their late-2022 peak, while workers aged 30-plus in the same fields grew 6% to 12%. Aggregate AI-exposed employment across all ages fell just 0.2% over the year, masking the generational split. The declines concentrate where AI substitutes rather than complements.
Goldman Sachs (US Daily note by economist Elsie Peng, 6 April 2026): AI substitution cut roughly 25,000 US jobs a month over the past year and raised unemployment 0.16 percentage points, while augmentation added about 9,000 a month, for a net drag of 16,000 jobs a month and a 0.1pp rise in unemployment; “these negative effects fall largely on less experienced workers,” hitting Gen Z hardest.

Roles seeing rising hiring:

AI and machine-learning specialists, big-data specialists, fintech engineers, security-management specialists (WEF fastest-growing list).
Data-centre trades and operations: electricians, pipefitters, ironworkers; security roles related to data-centre operations hit 66,800 postings in 2025, up 124% year on year.
Senior, AI-fluent software roles: Indeed Hiring Lab found US software-development postings rose almost 15% since Claude Code’s late-February 2025 launch while overall postings fell 7%, but 71% of that increase came from senior roles and 37% from postings mentioning AI in the title.

Where exposure and measured change diverge: PwC’s paradox is the sharpest illustration: the least AI-exposed occupations are growing employment far faster than the most exposed, even as AI-exposed industries show around 4x productivity growth and higher pay. High applicability (Microsoft) and high usage (Anthropic) do not map neatly onto job losses, because augmentation, new demand and quality limits intervene.

6. The new graduate trying to get a job right now

Verdict: this is the hardest entry-level white-collar market in over a decade, and computing graduates are unusually exposed.

US: NY Fed data show recent-graduate (22 to 27) unemployment at about 5.6% in Q2 2026 with underemployment at 42%, against about 3.1% for all college graduates. Computer science recent-grad unemployment is about 6.1% and computer engineering 7.5% to 7.8%, higher than several non-STEM majors such as art history (about 3%). Caveat: the NY Fed major-level figures come from the 2023 American Community Survey with wide confidence intervals (computer engineering’s 7.5% could plausibly sit between 4% and 11%), so the “CS is riskier than art history” framing is statistically fragile.
Postings: Indeed’s software-development posting index sat around 73 as of April 2026 (base 100 in February 2020); tech postings were down 36% from February 2020 levels by mid-2025, with software-engineer postings down 49%. Entry-level tech hiring dropped about 25% year on year in 2024.
Mechanism: Stanford finds the entry-level decline is driven by stalled hiring, not layoffs; junior employment at AI-adopting firms drops 9% to 10% within six quarters while senior employment barely moves.
India: TCS announced about 12,000 job cuts in July 2025, the largest in Indian IT history, framed as AI-driven “workforce rationalisation” by Nasscom. But fresher hiring is recovering: top IT firms expect to onboard about 82,000 graduates in FY2026, and Infosys said it would hire 20,000 freshers in 2025 while reporting 5% to 15% software-development productivity gains from AI. IIT and NIT placement rates weakened even in CSE (ranging roughly 25% to 92% across premier institutes for the 2023 to 2024 batch, with many IITs at 50% to 60% overall).

Practical implication: Graduates need demonstrable applied-AI portfolios (shipped projects, agent orchestration, GitHub history), because employers are concentrating scarce hiring on AI-fluent and senior-capable candidates. A degree alone no longer clears the bar.

7. The current student deciding what to study

Verdict: keep studying computing, but specialise toward AI-augmented, judgement-heavy work, not routine coding.

Enrolment is diverging by country. In the US, CS enrolment is falling for the first time in nearly two decades: the CRA Taulbee Survey (55th edition, 2025 data) found bachelor’s CS enrolment down 3.1% (a swing from +6.6% the year before), and National Student Clearinghouse data show computer and information sciences undergraduate enrolment down about 8.1% in fall 2025 (roughly 659,700 to 606,100 students), the steepest decline of any field. A CRA CERP pulse survey (October 2025) found 62% of computing units reported declining undergraduate enrolment. In India, engineering and CS admissions are still growing: AICTE approved 14.90 lakh B.Tech seats for 2024 to 2025, of which 12.53 lakh filled (highest in eight years), with the all-India enrolment rate at a five-year high of 75.07% and demand shifting strongly toward CS, AI/ML, data science and IoT programmes. (Note: I could not verify the commonly cited “Forrester 20% CS enrolment decline”; Forrester’s actual 2026 prediction concerned deferred AI spending and doubled developer time-to-fill, not enrolment, so treat the 20% figure as unverified. The verified US declines are the 3.1% Taulbee and 8.1% NSC numbers.)
Skills employers now want: AI literacy, agentic AI (postings up over 280%), prompt engineering, plus deployment skills (Python, AWS, scalability, workflow management), signalling AI is moving into infrastructure and operations. WEF finds 39% of core skills will change by 2030 and analytical thinking, resilience and creative thinking top the human-skills list.
Premium shift: PwC finds skills sought in the most AI-exposed jobs are changing 66% faster, and AI-exposed entry roles are seven times more likely to require traditionally senior skills such as judgement and leadership.

Practical implication: The winning profile is “domain expertise plus AI”, not pure coding. Curricula should push agent orchestration, evaluation and verification, and AI-augmented domain work (finance, law, healthcare, design) where human judgement still gates the output.

8. The existing knowledge worker

Verdict: augmentation dominates displacement for now, and AI fluency is the single clearest driver of higher pay.

Wage premium: PwC’s Global AI Jobs Barometer put the AI-skills wage premium at 56% in 2025 (up from 25% the prior year) and 62% in 2026 (up from 57%), reaching as high as 118% in some sectors such as consumer markets and 16% in government and public-sector work. Headcount at the most AI-exposed firms grew 52% versus 36% at the least exposed (2018 baseline).
Productivity RCTs:
- Brynjolfsson, Li and Raymond (“Generative AI at Work”, NBER WP 31161, 2023; published QJE, May 2025), a staggered rollout to about 5,000 customer-support agents at a Fortune 500 firm: the AI assistant raised issues resolved per hour by 14% on average (15% in the published version), with a 34% gain for novice and low-skilled agents and minimal effect on experts, a “levelling up” effect.
- Noy and Zhang (Science, 14 July 2023), a preregistered experiment with 453 professionals on writing tasks: ChatGPT cut average completion time 40% and raised output quality 18%, compressing inequality between workers.
- Dell’Acqua et al. “jagged frontier” (Harvard/BCG, 758 consultants, September 2023, published in Organization Science 2025): inside AI’s frontier, consultants did 12.2% more tasks, 25.1% faster, with about 40% higher quality; outside the frontier, AI users were about 19 percentage points less accurate, over-relying on AI where it was weakest.
- METR (10 July 2025, arXiv 2507.09089), an RCT with 16 experienced open-source developers on 246 real tasks: developers were about 19% slower using early-2025 AI tools, despite expecting a 24% speed-up and still believing afterwards they had been about 20% faster, a caution against assuming senior productivity gains.
Task composition: Anthropic’s data shows a slow rise in automation’s share of tasks over time (directive use jumped from 27% to 39% in eight months), though its November 2025 sample showed augmentation back ahead at 52% to 45%.

Most at risk mid-career: routine information processing (insurance claims clerks, bill collectors, basic financial analysis, template writing, tier-one support). Least at risk: roles combining physical presence, high-stakes judgement, regulation, and relationships (lawyers, construction managers, physicians, per Goldman’s augmentation category).

Practical implication: Mid-career workers should move up the task ladder, from doing the task to directing, checking and integrating AI output, and build verifiable AI-augmentation track records.

9. Highest and lowest AI penetration by task and occupation

Highest-penetration tasks and occupations (from real usage data):

Software and coding tasks (Anthropic: computer and mathematical work is about a third of Claude.ai use, nearly half of API traffic).
Writing and content drafting (Microsoft: writing among the most common AI-performed activities).
Information gathering, summarising and research (Microsoft: gathering information and writing top the list).
Translation (Microsoft: translators have the highest applicability score).
Customer service and sales support (high applicability; heavy automation attempts).
Routine data and analytics work (Indeed: 45% of data and analytics postings mentioned AI by December 2025, the highest of any sector).

Lowest-penetration tasks and occupations, with the reason:

Physical and manual work (truck driving, warehouse, construction, care work): reason is physical embodiment, outside generative AI’s reach.
High-stakes clinical and legal judgement: reason is verification cost and liability (regulatory and high-stakes).
Relational and empathetic work (complex customer escalations, care): reason is relational, the exact gap that forced Klarna’s rehire.
Highest-paid strategic roles: Anthropic finds usage is lower at both the lowest and highest wage ends.

Supporting frontier evidence:

OpenAI’s GDPval (released 25 September 2025) tested frontier models on 1,320 real deliverables across 44 occupations in 9 industries (each contributing over 5% of US GDP). The best model, Claude Opus 4.1, produced work rated as good as or better than industry experts in just under half of tasks (about 49% win/tie), with GPT-5 around 40.6%. Performance “more than doubled” from GPT-4o (spring 2024) to GPT-5 (summer 2025). This shows models near parity on discrete deliverables but not on whole occupations.
MIT’s “GenAI Divide: State of AI in Business 2025” (Project NANDA, MIT Media Lab, released mid-2025) found about 95% of enterprise generative-AI pilots delivered no measurable P&L return, with only about 5% extracting significant value, attributing the gap to a “learning gap” (lack of memory and contextual adaptation) rather than model quality. This is an industry report, not peer-reviewed, and its dataset is unpublished.
The Anthropic Economic Index shows augmentation and automation running close (roughly 52% to 45% augmentation in the November 2025 sample), confirming that even where AI is used heavily, it collaborates about as often as it replaces.

Verdict: Penetration is highest in digital information work that is easy to verify and rich in training data, and lowest where work is physical, high-stakes, relational, regulated, or data-poor. The GDPval and MIT findings together show a gap between task-level capability (high) and whole-job or enterprise-value capture (still low).

What would change this assessment

If Challenger’s AI-cited layoffs break decisively above 10% of total cuts, or Goldman’s net monthly AI job loss climbs well beyond 16,000, the picture shifts toward genuine displacement.
If Stanford’s young-worker gap stops widening (it hit 19% in August 2026), the entry-level thesis weakens.
If GDPval win rates cross well above 50% for whole-occupation bundles (not just discrete tasks), whole-job displacement risk rises materially.
If the wage premium compresses, AI skills are commoditising and differentiation must move to domain depth.

Caveats

Measured versus projected: The large positive job-creation numbers (WEF 78 million net, 4.7 million data-centre construction jobs) are forecasts. The measured numbers (AI-cited layoffs about 71,825 cumulative; AI job-posting share about 2.5%; young-worker gap 19%) are smaller and more concrete.
Single-origin claims to flag: the 4.7 million data-centre construction jobs (American Edge Project, Meta-linked advocacy); the Klarna reversal narrative (largely one Bloomberg interview); Challenger’s AI-cited layoffs (one national tracker, self-reported reasons); and Goldman’s 16,000-a-month figure (one modelled estimate combining exposure and IMF complementarity indices).
Conflicting signals: Anthropic’s own economist rebutted its CEO; Indeed shows AI-exposed postings first falling then rebounding; PwC shows least-exposed jobs growing fastest despite AI creating value in exposed ones. AI’s labour effect in 2026 is real, concentrated at the entry level, but not the economy-wide displacement wave the loudest 2025 predictions implied.
India data is thinner and often routed through education-media outlets rather than primary AICTE or Nasscom documents.
NY Fed major-level unemployment figures carry wide confidence intervals and lag (2023 ACS base), so precise CS-versus-other-major comparisons should be treated cautiously.
MIT GenAI Divide is a non-peer-reviewed industry report with an unpublished dataset; its 95% figure has drawn methodological criticism and should be read as indicative, not definitive.