SASIGNAL ATLASCross-industry intelligence / Research desk
SIGNAL ATLAS / RESEARCH DESK

Education, edtech & workforce development

Dossier · Education, edtech & workforce development · Original Phase 1 research

Education, edtech & workforce development

Industry ID: 22 | Slug: education-workforce | Researched: 2026-09-15 | Analyst: agent

Method note (read first). WebSearch budget for the programme was exhausted before this sector ran (200/200 calls consumed; one confirmatory attempt returned the budget message). All evidence below was gathered by direct WebFetch on primary-source URLs per Addendum 2 of the macro brief. Tier-A verification is therefore sound but discovery is narrow: I could only verify what I could guess the URL for. Sector 22 should be queued for re-run at full search budget. See §13.

1. Definition and boundaries

In scope. The production, financing, delivery, measurement and certification of human capability, across the whole life course:

  • K-12 schooling — public and private, curriculum, instructional materials, assessment.
  • Higher education — public and private universities, community colleges, their finance, enrolment and labour force.
  • Edtech — software and services sold into education institutions or direct to learners.
  • Corporate learning and development — enterprise training, upskilling, reskilling.
  • Credentialing and assessment — degrees, certificates, licensure, testing, verification.
  • Labour-market reskilling and transition — public workforce systems, apprenticeships, bootcamps, and the measurement of how education translates into employment.

The last item is the sector's most consequential boundary. The labour-market half is treated here as first-class. This is the sector where the AI-and-jobs question becomes measurable, because the entry-level graduate cohort is the cleanest natural experiment available: it is large, well-instrumented (NY Fed, BLS, ADP), and consists of workers whose tasks are disproportionately the ones generative AI does well.

Explicitly out of scope, and who owns it:

Excluded Owned by
The foundation models themselves (training, inference, model capability) Sector 01 — AI & foundation models
General HR software, payroll, ATS, core HCM Sector 02 — enterprise software
Childcare and early-years custodial care (as care, not as instruction) Healthcare / social assistance
Student housing, campus construction, university real-estate portfolios Sector 18 — real estate & construction
Student lending as a credit asset class (securitisation, servicing economics) Sector 07 — finance & fintech
Consumer language/skill apps competing on entertainment mechanics Contested — see below

Boundary disputes worth naming.

  1. Duolingo is the live definitional fight. Its DAU grew 23% year over year in Q2 2026 (Duolingo Q2 2026 release). Is that education, or is it a consumer engagement app with education-shaped content? Its retention mechanics, monetisation and cohort behaviour look like sector 17 (social/creator apps). Classify by what it is bought on: streak maintenance is entertainment, certification is education. Most analysts double-count it.
  2. Corporate L&D vs enterprise software. Coursera's Enterprise line and Udemy Business are sold to CHROs alongside HCM suites. The buying centre is HR, not academics.
  3. Is "AI literacy training" education or is it AI-sector demand generation? Much of what is counted as reskilling spend is vendor-funded enablement for that vendor's own product. This is a genuine measurement problem, not a quibble — see §10.
  4. Assessment is quietly becoming a security industry. Proctoring, identity verification and provenance detection have more in common with sector 04 (cybersecurity) than with pedagogy. The margin is migrating with it.

2. Subcategories

  1. K-12 instruction and curriculum — publishers, core and supplemental materials, and increasingly teacher-productivity AI. Buyer is the district; the purchase is political.
  2. K-12 operations software — SIS, LMS, assessment platforms. Sticky, low-growth, heavily consolidated into private equity ownership.
  3. Higher education institutions — the enrolment, tuition and endowment business. Distinguished by extreme cost structure rigidity: ~55-60% of spend is staff.
  4. Higher education services — OPMs, enrolment marketing, student success software. Distinguished by being the part of higher ed that can be disrupted quickly, and has been.
  5. Consumer learning — direct-to-learner subscriptions (Chegg, Coursera Consumer, Duolingo). Distinguished by being the first subsegment that generative AI measurably broke.
  6. Corporate L&D and skilling platforms — enterprise seats, content libraries, skills taxonomies. Distinguished by buying on compliance and retention, not learning outcomes.
  7. Credentialing and assessment — degrees, industry certificates, licensure exams, admissions testing. Distinguished by regulatory moats and network effects in recognition.
  8. Workforce development and public reskilling — WIOA-funded systems, community college workforce divisions, apprenticeships, and now Workforce Pell. Distinguished by being funded by government and judged on placement, not completion.
  9. Labour-market data and matching — job-posting analytics, skills graphs, graduate outcome tracking. Distinguished by selling intelligence about the sector rather than education itself. Structurally the highest-margin niche.
  10. International student mobility — recruitment, pathway programmes, visa services. Distinguished by being a cross-border revenue business dressed as an academic one.
  11. Early-years and supplemental tutoring — private tutoring, test prep, after-school. Distinguished by cash-pay consumer economics and cultural concentration in East Asia.
  12. Faculty and teacher labour — the employment side: pay, pipeline, vacancies, unions. Distinguished by being the sector's largest cost and its binding capacity constraint.

3. Market structure

Concentration: barbell, not oligopoly. The sector is simultaneously one of the most fragmented and one of the most concentrated markets in the economy, depending on layer.

  • Delivery is atomised. ~13,000 US school districts and several thousand degree-granting institutions, each an independent procurement authority. No vendor holds pricing power over the buyer set as a whole.
  • Infrastructure is concentrated. SIS and LMS layers have consolidated into a handful of private-equity-owned platforms. Assessment and credentialing are near-monopolies in their niches (licensure exams, admissions testing).
  • Consumer learning is winner-take-most and has just turned over. Chegg held effective monopoly economics in US homework help; its revenue fell 51% year over year in Q2 2026 to $51.8m from $105.1m (Chegg Q2 2026 results, 2026-08-06). That is not competitive erosion, it is substitution by a general-purpose product.

Where margin actually sits. Not in teaching. Margin concentrates in three places:

  1. Credential recognition. The right to say a credential counts. Regulatory and network moats, near-zero marginal cost, and no substitution pressure from AI — an AI can do the work of a certificate holder but cannot issue the certificate.
  2. Compliance-mandated software. Anything a district or institution must buy to satisfy a statute or accreditor. Demand is inelastic and renewal is near-automatic.
  3. Enterprise seat aggregation. Coursera's Q2 2026 adjusted EBITDA margin reached 14.3% from 9.6% a year earlier (Coursera Q2 2026 results), largely through consolidation rather than pricing.

Margin does not sit in content. Content has been commoditised twice — first by OER, then decisively by generative models. Chegg's residual profitability comes from Skilling ($17.5m, +2% y/y), not from its historic content library.

Barriers to entry. Low technically, very high commercially. Anyone can build a tutor; almost no one can get a district to buy it. The real barriers are procurement cycles measured in school years, accreditation, student-data privacy law, and the fact that the buyer (administrator), user (teacher) and beneficiary (student) are three different people with misaligned incentives. This is why MagicSchool's distribution is the notable fact about it — approximately 8 million educators registered and "one in five children in America go to a school that is in partnership with MagicSchool" (Crunchbase News, 2026-08-05) — on a total disclosed funding of only ~$63m. Distribution, not model quality, was the scarce asset.

Who has pricing power. Licensure and admissions-test bodies; accreditors; a small number of compliance software vendors. Nobody else. Institutions have lost pricing power: Gallup records the share of Americans saying a college education is "very important" falling from 75% in 2010 to 70% in 2013 to 51% in 2019 to 35% in 2025 (Gallup, 2025-09-11).

Market size. I will not launder a market-size estimate. The reliable public anchors are flows, not stocks:

  • Edtech venture funding: ~$2.8bn globally in 2025 year-to-date, "roughly flat with 2024"; US ~$1.2bn, "roughly on par with 2023" (Crunchbase News, Joanna Glasner, 2025-11-21). Label: fact for the tracker's own count; note Crunchbase's known undercount of undisclosed rounds. For scale: this is roughly 0.5% of H1 2026 global VC ($510bn Crunchbase / $560.4bn KPMG per the macro brief). Edtech is a rounding error in capital terms.
  • US education/training/library occupations: ~871,900 projected annual openings, median annual wage $60,570 (May 2025) (BLS OOH, updated 2026-08-27). The sector is enormous as employment and trivial as venture capital. That asymmetry is the single most important structural fact about it.

4. Who matters

Leading companies

Notable startups

Active investors

Platforms and standards bodies

  • 1EdTech (formerly IMS Global) — https://www.1edtech.org — LTI, Caliper, Comprehensive Learner Record; the interoperability layer that determines integration cost.
  • Open Badges / Comprehensive Learner Record — credential portability standards.
  • ISTE — https://iste.org — ed-tech practice standards.

Regulators and government agencies

  • US Department of Education — https://www.ed.gov — Secretary Linda McMahon.
  • NCES (IES) — https://nces.ed.gov — IPEDS, NAEP, School Pulse Panel.
  • Bureau of Labor Statistics — https://www.bls.gov — CPS, JOLTS, Employment Projections, OOH.
  • SEVP / ICE — https://www.ice.gov/sevis — international student tracking.
  • US Department of State, Bureau of Consular Affairs — visa issuance.
  • Office for Students (England) — https://www.officeforstudents.org.uk — financial regulator.
  • Ofqual (England) — qualifications and examinations regulator.
  • Regional and programmatic accreditors — the effective gatekeepers of Title IV eligibility.

Research institutions

Trade organisations

Consumer and civil-society groups

  • Teachers' unions (NEA, AFT; UCU in the UK) — the binding constraint on labour-cost change.
  • Lumina Foundation — https://www.luminafoundation.org — attainment goals, Gallup partnership.
  • Common Sense Media — child online safety and AI in schools.

5. Products, business models, technologies, customers

Major products. Instructional materials; learning management and student information systems; assessment and proctoring; tutoring and homework help; degree and certificate programmes; enterprise course libraries; skills taxonomies; graduate outcome analytics.

How money is actually made today.

  • Tuition and public appropriation — the overwhelming majority of sector revenue, and the least discussed by technology analysts because it is not a "product".
  • Per-student-per-year site licences — the K-12 software standard. Priced on enrolment, which means the demographic cliff is a direct revenue haircut with a one-year lag.
  • Consumer subscription — the model generative AI broke. Chegg is the proof.
  • Enterprise seats — Coursera Enterprise reached $140.0m in Q2 2026, +118% y/y, but that growth is primarily the Udemy acquisition, not organic demand.
  • Credential and exam fees — the highest-margin line in the sector.
  • International student premium tuition — cross-subsidises domestic teaching and research in the US, UK, Canada and Australia. A revenue line that is a foreign-policy variable.

How it is changing. Three shifts are underway simultaneously:

  1. From content to workflow. Content is free now. The products that grew are the ones that sit in a teacher's or administrator's workflow (MagicSchool) rather than in a student's study session (Chegg).
  2. From completion to placement. Workforce Pell conditions federal money on "strong earnings, job placement, and completion benchmarks" (US ED, 2026-09-14). Outcome-linked funding is arriving in statute, not just in rhetoric.
  3. From assessment-of-product to assessment-of-process. Take-home written work no longer evidences anything. The institutional response is migrating spend toward supervised, oral and in-person assessment — which is more labour-intensive, not less. AI raises the cost of assessment even as it lowers the cost of instruction.

Technologies that matter. Generative models in teacher-facing workflow tools; retrieval over curriculum corpora; identity verification and proctoring; learner-record interoperability (1EdTech LTI/CLR); skills-graph inference from job postings and résumés; and — most underrated — administrative payroll data (ADP) as a near-real-time labour-market instrument, which is what made the entry-level finding in §8 possible at all.

Customers and what they buy on.

  • Districts buy on political defensibility, procurement compliance and teacher retention. Not learning gains. Efficacy evidence is a tiebreaker at best.
  • Universities buy on enrolment yield and accreditation risk.
  • Enterprises buy L&D on retention, compliance and internal-mobility narrative.
  • Learners buy on credential recognition and, increasingly, on price — because the perceived value of the credential itself has fallen (Gallup: 35%, §3).
  • Governments buy on placement and earnings, increasingly measured.

6. Geography

Demand is global and demographically determined. It is falling in the rich world's traditional-age cohort and rising in South Asia and sub-Saharan Africa. The US high-school graduate cohort peaked in 2025 and is projected to fall 13% through 2041, with 38 states declining (WICHE, Knocking at the College Door, 11th ed., 2024).

Capital concentrates overwhelmingly in the US — but the more important geographic fact is how little capital this sector attracts anywhere. Global edtech venture funding of ~$2.8bn (2025 YTD) against $510-560bn of global VC means the sector is capital-starved in every region.

Regulation is the most geographically fragmented layer. US: federal Title IV eligibility plus 50 state K-12 regimes plus private accreditors. EU: national competence for education, but the AI Act now bites on employment-context AI — and note from the macro brief that high-risk obligations for employment AI were deferred from 2026-08-02 to 2027-12-02. Any vendor claiming its hiring or assessment product is compliance-driven in the EU right now is selling against a deadline that did not bind.

Non-US market — the European Union (regional source: Eurostat). In 2025, 44.8% of EU 25-34-year-olds had completed tertiary education, against the Council's target of at least 45% by 2030 — "almost half the EU countries have already reached the target" (Eurostat, Educational attainment statistics, data extracted April 2026). The distribution is extreme: Ireland, Luxembourg, Lithuania and Cyprus at 60% or above; Romania below 25%. The gender gap is now the dominant structural feature: 50.6% of women versus 39.3% of men, an 11.3pp gap. And 85.7% of 20-24-year-olds hold at least upper secondary education.

Two implications Anglophone coverage routinely misses. First, the EU is about to hit a tertiary-attainment target that the US political conversation has abandoned as a goal entirely — the two blocs are moving in opposite directions of intent. Second, an 11.3pp gender gap in attainment is a labour-market composition shock arriving over the next two decades, and it is larger and better documented in the EU than the equivalent US discussion.

International mobility is where geography becomes a P&L line. The US hosted 1,177,766 international students in 2024/25, up 5% — but new enrolments fell 7% to 277,118, while OPT participation rose 21% to 294,253 (IIE Open Doors, 2025-11-17). India 363,019 (+10%), China 265,919 (−4%). See §8 for why the headline "+5%" is misleading.

Non-US data access is materially worse. UK HESA returned HTTP 403 through the proxy; the OECD Education at a Glance landing pages 404'd on every URL pattern attempted; UK study-visa detail pages 404'd. This is recorded in §13 and in data_gaps, and it means the non-US picture in this dossier rests on Eurostat and OfS publication metadata alone.

7. Historical trend patterns

This sector has the richest record of false positives of any sector in the programme. It is structurally prone to them: the buyer cannot measure the product, the press is staffed by graduates who find education intrinsically interesting, and every wave is morally attractive.

The 25-year record of predicted disruptions that did not happen on schedule:

  1. MOOCs (2012-2014). "The year of the MOOC." Universities were to be unbundled within a decade. Outcome: completion rates in low single digits; the surviving companies pivoted to enterprise seats and degree programmes — i.e. they became channel partners of the institutions they were to replace. Coursera's Degrees revenue is $13.4m in Q2 2026, down from $15.7m — fourteen years after the disruption was announced, the degree line is shrinking and is under 5% of company revenue. The institutions won.
  2. Coding bootcamps (2014-2019). Framed as a general-purpose replacement for the CS degree. Resolved by the 2022-2024 tech hiring contraction and then, decisively, by AI coding assistants removing the junior-developer rung the model depended on.
  3. 1:1 devices and "personalised learning" (2015-2019). Enormous capex, hardware delivered, learning gains not demonstrated at scale.
  4. The pandemic remote-learning boom (2020-2021). The sector's only genuine capital supercycle — and the cleanest reversal. Enrolment and usage collapsed on reopening; valuations followed. 2U/edX ended in bankruptcy; Chegg is down 51% y/y and describes itself as "rearchitecting the company to be AI-first." Anyone modelling 2026 edtech against 2021 comparables is modelling an artefact.
  5. "Skills-based hiring" / the death of the degree requirement (2017-present, recurring). Announced at least three times. Employers remove degree requirements from postings; actual hiring composition barely moves. This one is important because it is being announced again right now on AI grounds, and the prior two announcements did not verify.
  6. Learning analytics / "big data will personalise education" (2013-2017). Produced dashboards, not outcomes.

What has actually cycled, reliably:

  • Countercyclical enrolment. Community college and adult enrolment rise in recessions and fall in tight labour markets. This is the most dependable regularity in the sector, and it is a confound for every AI-displacement enrolment story.
  • Public funding cycles. State appropriations fall in recessions and are restored slowly; tuition rises to compensate; political backlash follows with a lag of roughly one cycle.
  • Assessment integrity panics. Calculators, the internet, Wikipedia, essay mills, now generative AI. Each was called the end of assessment. Each ended with assessment changing format rather than disappearing. That history is the strongest argument for treating the current integrity crisis as real-but-survivable — with one distinction noted in §8.
  • Demographic determinism. Birth cohorts are the sector's tide. Eighteen years of advance notice, and institutions still fail to act on it. This is the highest-confidence, lowest-uncertainty forecast available anywhere in the trend programme.

The lesson for 2026. In this sector, technology adoption by students runs years ahead of adoption by institutions, and revenue follows institutions. Pew finds 54% of US teens have used chatbots for schoolwork; MagicSchool has ~8m registered educators; and yet global edtech venture funding is flat at ~$2.8bn. Usage and capital have decoupled completely. The correct inference is not "edtech is about to boom" — it is that general-purpose AI captured the value and the education-specific vendors did not.

8. What is changing now (as of 2026-09-15)

8.1 The entry-level question — what the data actually shows

This is the most consequential and most misreported question in the sector. Handled with the fact / signal / forecast discipline the task requires.

FACT — measured, dated, primary.

Measure Value Source Date
Unemployment, recent college graduates (22-27) ~5.6% NY Fed 2026:Q2
Underemployment, recent college graduates 42%, "edged up" NY Fed 2026:Q2
Unemployment, bachelor's+ age 25+ 2.7% (2.7% a year earlier — flat) BLS Table A-4 Aug 2026
Unemployment, age 20-24 7.1% (from 9.2% Aug 2025 — improved) BLS Table A-10 Aug 2026
Unemployment, age 25-34 4.3% (from 4.4%) BLS Table A-10 Aug 2026
Unemployment, age 16-19 14.1% (from 13.9%) BLS Table A-10 Aug 2026
Headline U-3 4.1%, unchanged BLS Employment Situation Aug 2026
Nonfarm payrolls +162,000 BLS Employment Situation Aug 2026
Local government education payrolls +42,000 BLS Employment Situation Aug 2026
Job openings 7.3m (rate 4.4%); hires 3.2%; quits 1.9%; layoffs 1.0% BLS JOLTS Jul 2026

SIGNAL — leading indicator, not outcome. Brynjolfsson, Chandar and Chen, Canaries in the Coal Mine? (Stanford Digital Economy Lab, revised 2026-08-12), using ADP administrative payroll data through June 2026, find that employment of workers aged 22-25 in AI-exposed occupations now stands 19% below where it would be had it kept pace with their less-exposed peers. Three qualifications are load-bearing and are usually stripped out in secondary coverage:

  • The effect operates "primarily through reduced hiring of young workers rather than increased separations" — firms are not firing juniors, they are not opening the req.
  • Declines concentrate where AI substitutes for human tasks; where it complements workers, employment is flat or rising. The effect is task-level, not technology-level.
  • The authors explicitly characterise the results as "early, descriptive indicators — canaries in the coal mine — rather than causal estimates." Anyone citing this as proof that AI caused graduate unemployment is misrepresenting the paper's own stated claim.

FORECAST — modelled, future-tense. BLS Employment Projections 2025-2035, released 2026-08-27, project total employment rising from 170.3m to 176.2m, +5.9m jobs, +3.5% over the decade — against +10.9% in 2015-25. That is a halving-and-then-some of projected decadal growth. BLS explicitly attributes part of this to AI, warning that "the use of generative AI software" may "limit demand for some jobs" in arts, design and office and administrative support. The projected declines are concentrated exactly where an entry-level white-collar career used to start:

Occupation Change 2025-35 %
Cashiers −200,600 −6.5%
Office clerks, general −156,200 −6.0%
Customer service representatives −141,800 −5.3%
Secretaries and administrative assistants −114,100 −6.0%
Bookkeeping, accounting and auditing clerks −85,600 −5.6%
Data entry keyers −33,600 −25.5%
Payroll and timekeeping clerks −25,400 −15.9%
Word processors and typists −13,900 −34.4%

Meanwhile the largest growth is almost entirely non-graduate, in-person work: home health and personal care aides +847,300 (+18.1%), stockers and order fillers +250,700, fast food and counter workers +223,300, registered nurses +194,700.

The synthesis, stated carefully. There is a genuine and important tension in the evidence, and it should not be resolved by picking a side:

  • At occupation level, young workers in AI-exposed roles are measurably behind a reasonable counterfactual (−19%, ADP).
  • At age-cohort level, young workers in aggregate are doing better than a year ago (20-24 unemployment 9.2% → 7.1%).
  • At education level, the degree premium in unemployment terms is intact and stable (bachelor's+ 2.7%, flat y/y) — while recent graduates specifically sit at 5.6% with 42% underemployment.

The reconciliation that fits all three: this is a composition and entry-rate story, not a graduate-unemployment story. Young graduates appear to be reallocating out of AI-exposed occupations into less-exposed ones — which shows up as a large relative decline within exposed occupations, a rising underemployment rate (42%: graduates taking jobs that do not require the degree), and no deterioration in the aggregate age-group unemployment rate. The cost is being paid in match quality and early-career wage trajectory, not in joblessness.

That distinction matters enormously for decisions. "Graduates can't get jobs" implies a demand-stimulus or credential problem. "Graduates get jobs but not the jobs that build careers" implies a progression problem — and progression problems compound for a decade before they show up in any headline statistic. Underemployment at 42%, not unemployment at 5.6%, is the number to watch.

What would falsify the benign reading: a sustained rise in the 22-27 unemployment rate above ~6.5% while aggregate U-3 holds near 4%, or underemployment moving above ~45%.

8.2 The demographic cliff arrives

The US high-school graduate cohort peaked in 2025. Projections show −13% through 2041, with 38 states declining (WICHE, 11th ed.). The Midwest and Northeast are already falling.

The sector's most-cited counter-evidence is stale and should be retired: NCES's Condition of Education indicator still projects total undergraduate enrolment rising 9% to 16.8 million by 2031 — but that indicator was published in May 2023 and rests on a pre-revision pipeline. It contradicts WICHE directly. Recorded as a contradiction, not silently resolved.

Recent enrolment has been up, which confuses the picture: National Student Clearinghouse found total postsecondary enrolment +3.2% (+562,000) in spring 2025, undergraduate +3.5% to 15.3m (still −2.4% against pre-pandemic), community colleges +5.4% (+288,000), and undergraduate certificates +4.8%, now 20% above 2020. The growth is real but it is composition change — cheaper, shorter, more vocational — layered on top of a shrinking traditional-age pipeline. Certificates and associate degrees are growing; bachelor's degrees grew only 2.1% and remain below spring 2020.

Institutional consequence: closures. SHEEO's college-closures research cites a Chronicle analysis that over a five-year period an average of 20 campuses closed each month, affecting around 500,000 students — "mostly working adults, low-income students, and students of colour." SHEEO's own study of July 2004-June 2020 found over 100,000 students experienced an abrupt campus closure. Note both figures are pre-2021; I could not verify a current closure count. That is a significant gap given closures are the most decision-relevant downstream consequence of the cliff.

8.3 International students: a stock-versus-flow trap

The Open Doors headline — 1,177,766 students, +5% — is a stock number inflated by a flow that is not enrolment. OPT participation rose 21% to 294,253, and OPT participants are post-study workers counted in the total. The enrolment flow that actually drives next year's tuition revenue, new enrolments, fell 7% to 277,118.

For a tuition-dependent institution, new enrolments are the forward-looking variable and the total is a lagging one. Any institution or investor reading "+5%" as a demand signal is reading the wrong number. China is already declining (−4%); India (+10%) is carrying the total and is the concentration risk.

Policy risk is live but I could not verify current-year specifics: US State Department monthly visa issuance data stops at September 2025 on the public page, and UK study-visa detail pages 404'd. The UK Office for Students published Financial sustainability of higher education providers in England on 2026-05-14, but the figures are inside a PDF I could not retrieve. Recorded as gaps.

8.4 Assessment integrity has collapsed, and the response is costly

Pew (field 2025-09-25 to 2025-10-09, n=1,458 US teens 13-17, ±3.3pp; published 2026-02-24): 64% of teens use AI chatbots; 54% have used them for schoolwork; 10% say they do all or most of their schoolwork with chatbot help. On integrity: 59% think students at their school use AI to cheat at least sometimes, and 33% say it happens "extremely or very often." Among teens who have used chatbots for schoolwork, 76% think cheating is regular.

Anthropic's own analysis of 1m higher-education conversations (vendor-published, 2025-04-08 — label marketing/vendor analysis, not independent) found ~47% of conversations were "direct" (seeking output with minimal engagement), that Claude was doing Creating (39.8%) and Analyzing (30.2%) on Bloom's taxonomy — i.e. the higher-order work assessment is supposed to measure — and flagged conversations requesting rewrites to "avoid plagiarism detection." A vendor disclosing this against its own interest raises its credibility, but it is still one vendor's telemetry, not a population estimate.

The important structural point. Previous integrity panics (calculators, Wikipedia) were survivable because the technology could not produce the assessed artefact. Generative AI can. The unsupervised written artefact — essay, problem set, take-home exam, coursework — no longer carries information about the student. The response is a migration to supervised, oral, in-person and process-based assessment, all of which are more labour-intensive per student. This is the sector's central cost paradox: AI reduces the marginal cost of instruction and raises the marginal cost of assessment, and assessment is what institutions actually sell.

8.5 AI in the classroom is a teacher-productivity market, not a tutor market

The adoption that has scale is teacher-facing. MagicSchool: ~8m registered educators, "one in five children in America go to a school that is in partnership with MagicSchool," 3x y/y revenue growth, on ~$63m raised (Crunchbase News, 2026-08-05). Gallup research referenced in its own reporting found 60% of teachers use AI tools and 30% use them weekly (June 2025) — recorded as single source, since I could not retrieve the primary Gallup release.

Student-facing AI tutoring has enormous usage (Pew, above) and essentially no published population-level learning-outcome evidence. The gap between those two facts is the sector's largest unexploited research opportunity and its largest overclaim risk.

8.6 K-12 outcomes are deteriorating on the measures that matter

The Department of Education states that "twelfth graders today have the lowest reading scores ever recorded" and that "nearly 70% of 4th and 8th graders cannot read proficiently" (ED, 2026-09-09). NAEP's 2025 Long-Term Trend release reports 14% of 13-year-olds read for fun almost every day, down from 27% in 2012. Note the ED release is a policy communication with an explicit administration frame; the underlying NAEP data is Tier A but the framing is not neutral. Federal response is grant-based: EIR ~$500m/49 grants over two years, CLSD $256m to 24 states, SEED ~$90m, TQP ~$70m, IAL >$16m.

Public opinion has tracked the outcomes. Gallup: record-low 35% of Americans satisfied with K-12 quality (43% in 2024); 73% say education is going in the wrong direction; only 21% rate schools excellent/good at preparing students for current jobs. Yet 74% of parents are satisfied with their own child's school — the classic "my school is fine, the system is broken" pattern, which is why systemic reform has no constituency.

8.7 Credentialing policy actually moved — and adoption is negligible so far

Workforce Pell is live: created by the Working Families Tax Cuts Act, final rule published 2026-05-18, effective 2026-07-01. It extends Pell to short-term programmes as short as 8 weeks, conditioned on "strong earnings, job placement, and completion benchmarks," operation for at least a year, and approval by both the state Governor and the US Secretary of Education.

The adoption data is the story. As of 2026-09-14, three states and five programmes have been approved: Iowa (1 programme, 2026-08-04), Indiana (1, 2026-08-19), Nebraska (3, 2026-09-14 — phlebotomy technician, pharmacy technician, and CompTIA Tech+ with Google IT Support, all at Metropolitan Community College). This is the most significant change to US federal student aid eligibility in a generation, and after ten weeks it has produced five programmes at what appears to be a handful of community colleges. The dual-approval requirement is the throttle. Treat any 2026 forecast of Workforce Pell volume as unsupported.

8.8 Edtech capital: depressed, consolidating, and losing the value to general AI

Funding is flat at a low level (~$2.8bn global, ~$1.2bn US, 2025 YTD). The structural event of 2026 is consolidation, not investment: Coursera closed its Udemy acquisition on 2026-05-11. The combined entity's Q2 2026: revenue $299m (+60% y/y), Enterprise $140.0m (+118%), Consumer $158.6m (+29%), Degrees $13.4m (down from $15.7m); GAAP net loss $80.4m driven by $50.3m merger and integration costs plus $18.1m severance; non-GAAP net income $40.4m; adjusted EBITDA $42.7m at a 14.3% margin (from 9.6%); FY2026 guidance raised to $1.220-1.245bn; paid subscribers 1,655,000 (+44%).

And the number that undercuts the reskilling-boom narrative: combined enterprise customers 12,107, down 2% year over year. Revenue grew by acquisition; the customer base shrank.

Against that, Chegg: −51% y/y, guiding Q3 to $43-44m — below Q2 — with adjusted EBITDA collapsing to $1-2m from $9.1m. A company describing itself as "rearchitecting to be AI-first" while guiding revenue down 17% sequentially is managing decline, not transforming.

9. The five lists

Five most important current trends

  1. Entry-level white-collar hiring compression for recent graduates (T-22-01).
  2. The US demographic cliff and institutional consolidation (T-22-02).
  3. Collapse of unsupervised assessment integrity (T-22-03).
  4. Teacher-facing AI as the real classroom adoption path (T-22-04).
  5. Collapse in the perceived value of a degree (T-22-06).

Five fastest-growing signals

  1. Teacher-facing AI workflow tools at national scale (T-22-11) — ~8m educators.
  2. Short-term Workforce Pell credential financing (T-22-09) — fast in policy, slow in fact.
  3. Sub-degree credentials outgrowing bachelor's (T-22-14) — certificates +4.8%, +20% vs 2020.
  4. OPT as a share of international student presence (T-22-07) — +21%.
  5. Assessment re-architecture toward supervised and process-based formats (T-22-13).

Five trends most likely to affect businesses

  1. Entry-level hiring compression (T-22-01) — it is employers' choice, and it removes the training pipeline that produces mid-career staff five years out. Overlaps list 1.
  2. Corporate L&D reskilling spend and whether it is real (T-22-08).
  3. Edtech consolidation and vendor failure risk (T-22-05).
  4. Skills-based hiring claims versus measured hiring behaviour (T-22-10).
  5. International student flows as a labour-supply channel (T-22-07). Overlaps list 2.

Five trends most likely to affect consumers

  1. Degree value collapse and the household decision to enrol (T-22-06). Overlaps list 1.
  2. Assessment integrity collapse and what a credential now signals (T-22-03). Overlaps list 1.
  3. Institutional closures stranding students mid-programme (T-22-02). Overlaps list 1.
  4. AI chatbots as the default homework tool (T-22-04 / T-22-20).
  5. K-12 outcome deterioration, especially literacy (T-22-17 counterpart in §10).

Overlap is substantial and deliberate. Trends 1, 2, 3 and 6 appear on multiple lists because this sector's defining feature is that the same shock hits households, institutions and employers through different transmission channels at different speeds.

10. Overhyped / overlooked / cooling / reversing

Most overhyped

1. "AI has already destroyed entry-level jobs" as a settled empirical claim. The evidence that exists is a relative decline within AI-exposed occupations from administrative payroll data whose own authors call it "descriptive, not causal." The contrary evidence is that 20-24 unemployment fell from 9.2% to 7.1% over the same year and bachelor's+ unemployment was flat at 2.7%. Both the "AI jobs apocalypse" and the "nothing is happening" camps are cherry-picking. The defensible claim is narrower and more useful: entry rates into exposed occupations have fallen and underemployment is at 42%; unemployment has not risen.

2. Universal AI personal tutors delivering Bloom's "2 sigma" gains. Usage is enormous (54% of teens). Published population-level learning-gain evidence is absent. Meanwhile 12th-grade reading scores are the lowest ever recorded during the exact period of maximum AI tutor availability. That does not prove AI tutors fail — the confounds are overwhelming — but it decisively falsifies any claim that they are already working at scale.

3. The corporate reskilling boom. The headline is Coursera Enterprise +118%. The substance is an acquisition, and combined enterprise customers fell 2%. The LinkedIn Workplace Learning Report (2025) does not report a budget-increase share at all — it reports that only 36% of organisations qualify as "career development champions." There is no public, triangulated evidence of a broad L&D budget expansion. Vendor-push, not buyer-pull.

Most overlooked

1. Underemployment (42%), not unemployment (5.6%), is the decision-relevant series. Almost all commentary anchors on the unemployment rate because it is the familiar number. The underemployment rate is where the damage is visible, it "edged up" in 2026:Q2, and it directly measures the match-quality problem that compounds into a decade of lost wage growth.

2. AI raises the cost of assessment while lowering the cost of instruction. Every model of AI's effect on education economics assumes cost deflation. Assessment is the part institutions actually sell, and the credible response — supervised, oral, in-person, process-based — is more labour-intensive per student. This inverts the standard margin thesis and almost nobody is modelling it.

3. The dual-approval throttle on Workforce Pell. A generational expansion of federal aid eligibility has produced five programmes in three states in ten weeks because it requires both a Governor and the Secretary to sign. The mechanism of the bottleneck is public, checkable and predictive — and entirely absent from the coverage, which treats the policy as though volume follows automatically from statute.

4. The Open Doors stock/flow trap. "+5% international students" is reported everywhere; "new enrolments −7%" is reported almost nowhere, and OPT's +21% is what makes the totals diverge. Institutions budget off the wrong one.

5. Public-market disclosure in this sector is disappearing. Instructure and PowerSchool are private-equity-owned; 2U is gone. Following the pattern flagged in Addendum 2 (EA's take-private), the K-12 infrastructure layer no longer files quarterly. Phase 2 should treat this as a data-availability trend in its own right.

Trends that appear to be cooling

  • Online degree programmes. Indicator that turned: Coursera Degrees revenue $13.4m, down from $15.7m. After fourteen years, the degree line shrinks.
  • The consumer homework-help subscription. Indicator: Chegg −51% y/y, guiding down sequentially to $43-44m.
  • Coding bootcamps and "learn to code." Indicator: Crunchbase notes capital moving away from "coding academies and teaching platforms" as automation tools mature; BLS projects the clerical and routine-cognitive entry tier shrinking outright.

Trends that may reverse

  • International student decline could reverse fast — the mechanism is a single policy instrument (visa processing and eligibility rules). It reversed downward within one administration; it can reverse upward the same way. Low persistence by construction.
  • The enrolment recovery could reverse into the cliff. Current +3.2% enrolment growth is partly countercyclical adult enrolment. If the labour market stays at 4.1% U-3, that support fades exactly as the 2025-peak cohort starts shrinking. Watch the ratio of certificate to bachelor's growth.
  • Entry-level hiring could re-expand. If the −19% relative decline reflects firms over-extrapolating AI capability (a well-documented pattern in every prior automation wave), the correction comes as a hiring snapback when firms discover the mid-career pipeline they stopped filling. Early indicator: JOLTS hires rate for professional and business services rising while the AI-exposure gradient flattens.
  • The degree premium could reverse the other way. If AI compresses the value of routine-cognitive work, the credential's unemployment premium (2.7% vs 4.4%) may hold or widen even as its perceived value (35%) keeps falling — a widening gap between what the public believes and what the labour market pays. That divergence is already visible.

11. Risks and major uncertainties

Sector-specific risks

  1. Tuition-dependence + demographic decline = correlated institutional failure. Closures are not independent events; they cluster by region and by market position.
  2. International revenue is a foreign-policy variable. A single visa rule change moves a multi-billion-dollar cross-subsidy for US, UK, Canadian and Australian institutions.
  3. Labour cost rigidity. 55-60% of institutional spend is staff, largely tenured or unionised. Revenue can fall faster than cost can adjust. This is the closure mechanism.
  4. Assessment-integrity failure devalues the credential itself — the sector's core asset and its only real pricing power.
  5. Vendor concentration risk in K-12 infrastructure under PE ownership with reduced disclosure: the buyer cannot assess counterparty risk.
  6. Capital starvation. ~$2.8bn of global venture funding cannot finance a re-platforming of a sector this large. In a sticky-inflation, hawkish-lean environment (macro brief), it will not get cheaper.
  7. Regulatory whiplash. The EU deferred employment-AI high-risk obligations to 2027-12-02; US federal education policy has changed direction sharply. Compliance roadmaps built on either are low-persistence.

Genuine unknowns — "we don't know" (answerable, not yet answered)

  • Whether AI tutoring produces learning gains at population scale. Answerable by RCT; the studies are small and vendor-adjacent.
  • The true current college closure rate. The data exists in IPEDS; nobody I could reach has published a current count.
  • Whether corporate L&D budgets are actually rising. Answerable by survey; no triangulated public figure exists.
  • Current-year international enrolment. SEVIS and visa data exist but were not publicly current at this research date.

"Nobody can know" (irreducible)

  • Whether the −19% entry-level effect is AI capability, macro caution, over-extrapolation by managers, or all three. No identification strategy currently separates them, and the authors say so.
  • Where AI capability plateaus, which determines whether the clerical tier shrinks 6% (BLS) or disappears.
  • Whether a generation that learned with chatbots develops different, worse, or merely differently-distributed capabilities. The cohort has not aged into measurement.

12. Scenarios to 2030

Base — "Compression without collapse." Entry-level hiring stays compressed in exposed occupations; graduates reallocate; underemployment sits in the low-to-mid 40s; aggregate unemployment stays near 4%. Enrolment shifts toward certificates and community colleges as the cliff bites. A steady drip of small private-college closures. Edtech stays capital-starved and consolidates further. Falsifiable early indicator: NY Fed recent-grad underemployment stays in the 40-45% band for four consecutive quarters while U-3 stays within 3.8-4.5%.

Upside — "Complementarity wins." The complement-versus-substitute split in the Stanford data widens in favour of complements; firms discover the mid-career pipeline gap and re-open junior hiring; AI genuinely raises teacher productivity and measured outcomes stabilise. Indicator: recent-grad underemployment falls below 40% for two consecutive quarters, and JOLTS hires rate rises above 3.5% with growth concentrated in professional services.

Downside — "Progression failure." Entry-level compression persists five years. A cohort reaches age 30 without the experience that produces mid-career supply. Wage scarring appears in the data around 2029-2030. Enrolment falls as households rationally price the degree lower (Gallup already at 35%), accelerating closures into a self-reinforcing loop. Indicator: recent-graduate unemployment above 6.5% while aggregate U-3 stays below 4.5% — a divergence that cannot be explained by the business cycle.

Disruption — "The credential decouples from the institution." Employer-recognised AI-assessed competency verification reaches critical mass; Workforce Pell scales to hundreds of programmes; enrolment moves decisively to sub-degree credentials. Indicator: Workforce Pell approvals exceed ~200 programmes across 20+ states by end-2027, and certificate enrolment growth exceeds bachelor's growth by more than 5pp for three consecutive terms. Note the first half of that test is currently failing badly: five programmes, three states.

Regulatory — "Assessment and employment AI get rules." EU high-risk employment-AI obligations bind 2027-12-02; assessment integrity becomes an accreditation requirement; proctoring and provenance verification become mandatory spend. Costs rise sector-wide; compliance vendors capture the margin. Indicator: an accreditor makes AI-resistant assessment an explicit standard, or the EU deadline survives a second deferral vote.

Failure — "Correlated institutional collapse." The cliff, an international-student policy shock and a funding squeeze land together. Closures move from a drip to a wave; students are stranded mid-programme at scale; public confidence (already 35%) falls further and takes political support for public funding with it. Indicator: more than three institutions with enrolment above 5,000 announcing closure or merger in a single academic year, or the OfS reporting a majority of English providers in deficit.

13. Data gaps and limitations

Discovery was structurally limited. The programme's WebSearch budget (200/200) was exhausted before sector 22 ran. One confirmatory call returned the budget message; everything in this dossier was retrieved by direct WebFetch on guessed or known primary URLs. This sector should be prioritised for re-run at full search budget. Specifically, I could not do open discovery on: current college-closure counts, US 2026 international-student policy changes, corporate L&D budget surveys, or AI-tutoring efficacy literature.

Sources that blocked or failed:

  • UK HESA (hesa.ac.uk/data-and-analysis/students) — HTTP 403 through the proxy. HESA is the authoritative UK HE dataset; its absence materially weakens §6.
  • OECD Education at a Glance — every URL pattern attempted returned 404. No OECD figure appears in this dossier as a result. This is the largest single gap.
  • ICE SEVIS by the Numbers PDF — HTTP 403.
  • US State Department international students page — HTTP 403; the monthly NIV issuance page is public but stops at September 2025, so no current F-1 issuance trend.
  • UK Home Office study-visa detail pages — 404 on the year-ending-June-2026 sub-pages.
  • IBO (ibo.org) — HTTP 403; no exam-board assessment-policy evidence retrieved.
  • NationsReportCard.gov report pages — redirect loops; only the LTT headline was obtained.
  • SEC browse-edgar and data.sec.gov via curl — proxy connect_rejected. WebFetch on /Archives/ worked, consistent with the Addendum 2 playbook.

Figures I could not verify and have therefore not asserted:

  • Any OECD tertiary attainment, teacher salary or spending-per-student figure.
  • Current (2025-26) US college closure counts. The only figures available are the Chronicle's "20 campuses per month / ~500,000 students" and SHEEO's July 2004-June 2020 study — both pre-2021. I have flagged them as dated rather than presenting them as current.
  • OfS Financial sustainability 2026 findings. Publication date confirmed (2026-05-14); the figures are inside a PDF that did not retrieve.
  • 2025/26 international enrolment. Open Doors 2024/25 (published 2025-11-17) is the newest.
  • Duolingo Q2 2026 revenue, bookings, paid subscribers and guidance — only DAU +23% y/y was retrievable from the press release; the detailed metrics are in a shareholder letter that did not fetch.
  • Exact publication date of Coursera's Q2 2026 results release. The document was retrieved from investor.coursera.com under a /2026/ path and reports the quarter ended 30 June 2026; no printed release date was visible. Source records use the period end (2026-06-30) with this caveat rather than a guessed date.
  • NAEP 2025 Long-Term Trend scale scores. Only the "14% read for fun" datapoint and ED's characterisation ("lowest 12th-grade reading scores ever recorded") were obtained.
  • Gallup's "60% of teachers use AI, 30% weekly" (June 2025) — obtained second-hand from a later Gallup article. Single source; not independently verified.
  • Edtech venture funding peak-year figure (commonly cited as 2021). Crunchbase's article gave 2025 YTD and 2024/2023 comparisons but no peak figure. I have therefore not stated a peak-to-trough decline percentage anywhere in this dossier.
  • LinkedIn Workplace Learning Report: the 2026 edition could not be located; figures cited are from the 2025 edition (n=937 L&D/HR professionals, 679 learners).

Conflicting figures recorded rather than resolved: see §8.1 (ADP vs CPS), §8.2 (NCES vs WICHE), §8.3 (Open Doors stock vs flow), §8.8 (Coursera revenue vs customer count). Also carried from the macro brief: core CPI 2.4% vs core PCE ~3.3% (Aug 2026), unreconciled.

Known bias in retained sources. The US Department of Education releases used in §8.6 are government primary sources but carry explicit administration framing; I have used their numbers and flagged their framing. Anthropic's education report is vendor telemetry from a company that is also a party to this sector's disruption; labelled accordingly. Gallup and Pew are survey organisations with transparent methodology — treated as Tier A for their own survey results only.

14. Ranking scorecard

# Criterion Score Justification
1 speed_of_change 3 Student behaviour changed in months; institutional procurement, enrolment and demography move on multi-year to decadal cycles. Genuinely mixed, not fast.
2 economic_importance 5 ~871,900 annual openings in education occupations alone, and the sector supplies labour to every other sector; public spend is a major GDP component.
3 capital_invested 2 ~$2.8bn global edtech VC in 2025 YTD — roughly 0.5% of H1 2026 global VC. Capital-starved by any comparison.
4 company_product_density 4 Thousands of vendors across 13,000+ districts and several thousand institutions; highly fragmented, many distinct products to track.
5 regulatory_impact 4 Title IV, accreditation, Workforce Pell, visa policy, 50 state K-12 regimes, EU AI Act employment provisions. Rule-making determines most outcomes.
6 consumer_impact 5 Touches essentially every household directly, repeatedly, across the life course.
7 strategic_importance 4 National human-capital formation and the AI-skills transition; not physical infrastructure, so not a 5.
8 intelligence_demand 4 The AI-and-entry-level-jobs question is among the most-asked questions in the economy; demand is demonstrably high but buyers are budget-constrained.
9 paid_research_opportunity 3 An established research market exists, but the largest buyer class (institutions) is under financial stress and price-sensitive.
10 data_availability 5 Exceptional: IPEDS, NCES, NAEP, BLS CPS/JOLTS/EP, NY Fed, National Student Clearinghouse, Open Doors, Eurostat — mostly free, mostly structured, deep history. Among the best-instrumented sectors in the programme.
11 cross_industry_influence 4 Supplies labour to every sector, and the empirical evidence base on AI and employment originates from this sector's data.

15. Sources

  1. "The Labor Market for Recent College Graduates," Federal Reserve Bank of New York, https://www.newyorkfed.org/research/college-labor-market — data 2026:Q2, quarterly update (Aug 2026; exact page-update date not printed) — Tier A
  2. "The Employment Situation — August 2026," US Bureau of Labor Statistics, https://www.bls.gov/news.release/empsit.nr0.htm — 2026-09-04 — Tier A
  3. "Table A-4. Employment status of the civilian population 25 years and over by educational attainment," US BLS, https://www.bls.gov/news.release/empsit.t04.htm — 2026-09-04 — Tier A
  4. "Table A-10. Selected unemployment indicators, seasonally adjusted," US BLS, https://www.bls.gov/news.release/empsit.t10.htm — 2026-09-04 — Tier A
  5. "Job Openings and Labor Turnover — July 2026," US BLS, https://www.bls.gov/news.release/jolts.nr0.htm — 2026-09-01 — Tier A
  6. "Employment Projections — 2025-2035," US BLS, https://www.bls.gov/news.release/ecopro.nr0.htm — 2026-08-27 — Tier A
  7. "Occupations with the largest job declines, 2025-35," US BLS, https://www.bls.gov/emp/tables/occupations-largest-job-declines.htm — 2026-08-27 — Tier A
  8. "Occupations with the most job growth, 2025-35," US BLS, https://www.bls.gov/emp/tables/occupations-most-job-growth.htm — 2026-08-27 — Tier A
  9. "Education, Training, and Library Occupations," US BLS Occupational Outlook Handbook, https://www.bls.gov/ooh/education-training-and-library/home.htm — updated 2026-08-27 — Tier A
  10. Brynjolfsson, E., Chandar, B. & Chen, R., "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Stanford Digital Economy Lab, https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/ — revised 2026-08-12 — Tier A
  11. Lane, P., Falkenstern, C. & Bransberger, P., "Knocking at the College Door," 11th ed., WICHE, https://www.wiche.edu/knocking/ — 2024 — Tier A
  12. "Current Term Enrollment Estimates: Spring 2025," National Student Clearinghouse Research Center, https://nscresearchcenter.org/current-term-enrollment-estimates/ — 2025-05-22 — Tier A
  13. "Undergraduate Enrollment," Condition of Education, NCES, https://nces.ed.gov/programs/coe/indicator/cha/postsecondary-enrollment — May 2023 — Tier A (stale; see §8.2)
  14. "International Students," Open Doors, Institute of International Education, https://opendoorsdata.org/annual-release/international-students/ — 2025-11-17 (AY 2024/25) — Tier A
  15. "Educational attainment statistics," Eurostat Statistics Explained, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Educational_attainment_statistics — data extracted April 2026 (2025 reference year) — Tier A
  16. "U.S. Department of Education Approves Workforce Pell Grant Programs in Nebraska," US Department of Education, https://www.ed.gov/about/news/press-release/us-department-of-education-approves-workforce-pell-grant-programs-nebraska — 2026-09-14 — Tier A
  17. "Building Strong Readers: How the Trump Administration is Supercharging Evidence-Based Literacy to Combat Reading Crisis," US Department of Education, https://www.ed.gov/about/news/press-release/building-strong-readers-how-trump-administration-supercharging-evidence-based-literacy-combat-reading-crisis — 2026-09-09 — Tier A (government primary source; explicit administration framing)
  18. Chegg, Inc., "Q2 2026 Financial Results," Exhibit 99.1 to Form 8-K, SEC EDGAR, https://www.sec.gov/Archives/edgar/data/1364954/000136495426000085/a9901-financialresultsq220.htm — 2026-08-06 — Tier A
  19. Coursera, Inc., "Coursera Reports Second Quarter 2026 Financial Results," https://investor.coursera.com/news/news-details/2026/Coursera-Reports-Second-Quarter-2026-Financial-Results/default.aspx — quarter ended 2026-06-30 (release date not printed on retrieved page) — Tier A
  20. Duolingo, Inc., "Second Quarter 2026 Financial Results," https://investors.duolingo.com/static-files/2ef55449-ab04-4902-a5e3-01b13d67122a — Q2 2026 (webcast 2026-08-05) — Tier A
  21. "How Teens Use and View AI," Pew Research Center, https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/ — 2026-02-24 (field 2025-09-25 to 2025-10-09; n=1,458; ±3.3pp) — Tier A
  22. "Young adults in the U.S. are increasingly wary of AI, concerned it will take jobs," Pew Research Center, https://www.pewresearch.org/short-reads/2026/08/18/young-adults-in-the-us-are-increasingly-wary-of-ai-concerned-it-will-take-jobs/ — 2026-08-18 (field 2026-06-22 to 2026-06-28; n=3,488) — Tier A
  23. "Perceived Importance of College Hits New Low," Gallup, https://news.gallup.com/poll/695003/perceived-importance-college-hits-new-low.aspx — 2025-09-11 (field 2025-08-01 to 2025-08-20; n=1,094) — Tier A
  24. "Record-Low 35% in U.S. Satisfied With K-12 Education Quality," Gallup, https://news.gallup.com/poll/695174/record-low-satisfied-education-quality.aspx — 2025-09-16 — Tier A
  25. "Teachers Who Collaborate, Learn at Work Are More Satisfied," Gallup, https://news.gallup.com/poll/695492/teachers-collaborate-learn-work-satisfied.aspx — 2025-09-23 — Tier A
  26. "Anthropic Education Report: How University Students Use Claude," Anthropic, https://www.anthropic.com/news/anthropic-education-report-how-university-students-use-claude — 2025-04-08 — Tier C (vendor telemetry; self-published)
  27. Glasner, J., "Edtech-Specific Startup Funding Stays Low," Crunchbase News, https://news.crunchbase.com/venture/edtech-funding-stays-low/ — 2025-11-21 — Tier B
  28. Azevedo, M.A., "'Nobody Wanted to Give A Former Principal Money': How An Educator Built An Edtech AI Startup With $63M From VCs," Crunchbase News, https://news.crunchbase.com/venture/educator-built-edtech-startup-ai-magicschool-kahn/ — 2026-08-05 — Tier B
  29. "College Closures," SHEEO, https://sheeo.org/project/college-closures/ — Report Three, August 2023 (study period July 2004-June 2020) — Tier A (dated)
  30. "Financial sustainability of higher education providers in England 2026," Office for Students, https://www.officeforstudents.org.uk/publications/financial-sustainability-of-higher-education-providers-in-england-2026/ — 2026-05-14 — Tier A (publication metadata only; PDF not retrieved)
  31. "Workplace Learning Report 2025," LinkedIn Learning, https://learning.linkedin.com/resources/workplace-learning-report — 2025 (n=937 L&D/HR professionals; 679 learners) — Tier B (vendor-sponsored survey)
  32. "Introducing ChatGPT Edu," OpenAI, https://openai.com/index/introducing-chatgpt-edu/ — 2024-05-30 — Tier C (vendor announcement)
  33. "The Nation's Report Card — 2025 Long-Term Trend Assessment," NCES/NAEP, https://www.nationsreportcard.gov/ — 2025 assessment — Tier A (headline datapoint only; scale scores not retrieved)
  34. "School Pulse Panel," NCES, https://nces.ed.gov/surveys/spp/ — most recent collection February 2024 — Tier A
  35. "Immigration system statistics, year ending June 2026," UK Home Office, https://www.gov.uk/government/statistics/immigration-system-statistics-year-ending-june-2026 — 2026-08-27 — Tier A (landing page only; study-visa detail not retrieved)
Research provenance
Source artifact
02-dossiers/22-education-workforce.md
Corpus date
15 September 2026
Prepared for this site
16 September 2026
Site publication
18 September 2026
Verification
Inherited; not fully rechecked

More research in this sector · Explore trend records