A data-driven report (prepared December 26, 2025)
Executive summary
As we enter 2026, the evidence strongly suggests that AI is not “wiping out work” uniformly—but it is changing how work is done fast enough to reshape hiring, entry-level pathways, and productivity in many knowledge-heavy roles.
Key findings:
- Most “replacement” is task-level, not job-level. Multiple major research groups emphasize that current AI can theoretically automate large shares of work activities, but that’s not the same thing as entire occupations disappearing overnight. McKinsey notes today’s technologies could theoretically automate more than half of U.S. work hours, but stresses this is not a forecast of job losses and adoption takes time. McKinsey & Company
- Exposure is widespread; “fully automatable” jobs are rare. Penn Wharton estimates ~42% of U.S. jobs are “potentially exposed” (≥50% of activities could be automated), while only ~1% are “completely exposed” (could be performed entirely by AI without significant oversight). Penn Wharton Budget Model
- Global exposure is meaningful, but the highest-risk slice is small. The ILO’s refined global index finds about 1 in 4 workers are in occupations with some generative-AI exposure, while ~3.3% are in the highest exposure category. International Labour Organization
- Entry-level disruption is showing up in measurable data. A Stanford Digital Economy Lab working paper (updated November 2025) finds early‑career workers (ages 22–25) in the most AI‑exposed occupations saw a 13% relative decline in employment, even after controlling for firm-level shocks; impacts are concentrated where AI is more likely to automate rather than augment. Stanford Digital Economy Lab
- AI adoption is rising fast, but scaling remains uneven. McKinsey’s 2025 global survey reports 88% of respondents say their organizations use AI in at least one business function (up from 78% the prior year), yet only ~one-third report they’ve begun scaling AI programs; 23% report scaling an agentic AI system and 39% are experimenting. McKinsey & Company
- AI skills are being rewarded—strongly. PwC’s 2025 Global AI Jobs Barometer finds workers with AI skills command a 56% wage premium (up from 25% the year prior), and employer-sought skills are changing 66% faster in occupations most exposed to AI. PwC
- Training supply is lagging demand. OECD analysis finds one in three job vacancies has high AI exposure, but only 0.3%–5.5% of training courses in four studied countries deliver AI content, suggesting a major “skills bottleneck” going into 2026. OECD
Bottom line for 2026: AI is most likely to “replace jobs” through reduced hiring, role redesign, and attrition (especially in entry-level work) rather than mass immediate layoffs across the economy. The fastest changes will occur where tasks are text- or code-heavy, standardized, and easy to quality-check.
1) What “AI replacing jobs” actually means in 2026
Public debate often treats “replacement” as a binary (“job exists” vs “job gone”). In real labor markets, AI-driven replacement typically happens via four channels:
- Task substitution inside jobs (most common in 2024–2026)
- AI takes over part of a role (drafting, summarizing, simple customer inquiries, basic code scaffolding).
- The job title remains, but headcount needs may fall or skill requirements rise.
- Workflow redesign (replacement of processes, not roles)
- Organizations reconfigure entire workflows to exploit AI (fewer handoffs; fewer junior “prep” steps).
- McKinsey emphasizes redesigning workflows is key to achieving value at scale. McKinsey & Company
- Hiring shifts (especially entry-level)
- Firms stop backfilling junior roles because AI absorbs “starter tasks.”
- Stanford’s evidence points to disproportionate impacts on early-career workers in AI-exposed roles. Stanford Digital Economy Lab
- Full job elimination (least common near-term, but real in narrow pockets)
- Penn Wharton’s estimates imply fully AI-doable jobs exist, but they are a small share (~1%). Penn Wharton Budget Model
This is why credible research increasingly distinguishes between:
- Automation (AI can do the task end-to-end with minimal human involvement)
- Augmentation (AI boosts human output/quality/speed but humans remain central)
PwC uses this “automatable vs augmentable” framing explicitly and finds wages and job counts are still rising in many AI-exposed roles—suggesting more “reshaping” than wholesale elimination so far. PwC
2) Why 2026 is an inflection year
Three forces converging by 2026 increase the odds of visible job impacts:
A) AI capability is improving while costs fall
Stanford’s 2025 AI Index highlights that inference cost for systems performing at GPT‑3.5 level dropped 280×+ between Nov 2022 and Oct 2024, lowering the barrier to workplace deployment. Stanford HAI
B) Adoption is moving from “pilot” to “selective scale”
- McKinsey: 88% report AI use in at least one business function, but only ~1/3 have begun scaling programs. McKinsey & Company
- The near-term story for 2026 is less “everyone replaces jobs” and more “a meaningful minority reorganize work and start cutting unit labor requirements.”
C) “Agentic AI” pushes automation beyond single tasks
McKinsey reports 23% of respondents say they are scaling an agentic AI system and 39% are experimenting. McKinsey & Company
PwC also points to “agentic AI” as a near-term workforce multiplier. PwC
Agents matter for job replacement because they can:
- chain tasks together (research → draft → revise → submit → log results),
- interact across tools (email, CRM, ticketing systems),
- reduce the need for human “glue work,” often done by junior staff.
3) How many jobs are exposed in 2026 (and where)
3.1 Global view (ILO)
ILO’s refined global exposure index finds:
- ~1 in 4 workers are in occupations with some generative AI exposure
- ~3.3% are in the highest exposure category International Labour Organization
The ILO also finds exposure is higher in high-income countries and disproportionately affects women, reflecting occupational segregation and the concentration of cognitive/clerical tasks in certain roles. International Labour Organization
Interpretation for 2026:
Global “headline exposure” is large, but the highest-risk slice is much smaller. Policy and business planning should treat this as a targeting problem (which roles, which cohorts, which sectors) rather than a uniform labor shock.
3.2 OECD geography: cities vs non-urban regions
OECD’s city-network briefing note reports:
- ~1 in 3 workers in urban areas is exposed to generative AI
- vs ~1 in 5 in non-urban regions OECD
This reflects the higher concentration of:
- high-skill roles,
- professional services,
- education, ICT, and finance in cities—occupations more exposed to generative AI. OECD
Interpretation for 2026:
Expect metro labor markets to experience earlier and sharper “AI job churn,” while non-urban areas may feel slower direct effects from GenAI (though robotics and industrial automation can still matter there).
3.3 U.S. exposure and “fully replaceable” roles (Penn Wharton)
Penn Wharton estimates:
- ~42% of current U.S. jobs are potentially exposed (≥50% of activities could be automated by GenAI) Penn Wharton Budget Model
- 29% of jobs have no potential to substitute AI for workers Penn Wharton Budget Model
- ~1% of jobs are completely exposed (fully doable by AI without significant oversight) Penn Wharton Budget Model
It also estimates ~40% of labor income is potentially exposed to automation by GenAI (which matters for wage dynamics, bargaining, and distribution). Penn Wharton Budget Model
Interpretation for 2026:
The economy is not facing a near-term “50% unemployment” scenario. The more realistic near-term risk is:
- slower job growth in exposed occupations,
- weaker entry-level pipelines, and
- redistribution of wages toward AI-complementary skills.
4) What the labor market evidence says is happening already (as we enter 2026)
4.1 Entry-level employment declines in AI-exposed roles (Stanford)
Stanford Digital Economy Lab’s working paper (updated Nov 2025) finds:
- early-career workers (ages 22–25) in the most AI-exposed occupations experienced a 13% relative decline in employment, controlling for firm-level shocks Stanford Digital Economy Lab
- declines occur mainly through employment adjustments rather than compensation changes Stanford Digital Economy Lab
- impacts concentrate where AI is more likely to automate rather than augment human labor Stanford Digital Economy Lab
Why this matters for 2026:
This is one of the clearest large-scale signals that “AI replacement” can appear first as a career-entry bottleneck—not necessarily mass layoffs.
4.2 Employment stagnation where automation potential is highest (Penn Wharton)
Penn Wharton reports “suggestive evidence” that:
- job growth has stagnated in occupations with the most AI automation potential Penn Wharton Budget Model
- for jobs that could be performed entirely by GenAI, employment fell sharply in 2024 and was 0.75% lower than in 2021 (noting these jobs are only ~1% of total employment) Penn Wharton Budget Model
Interpretation for 2026:
Even if “fully AI-run jobs” are rare, they can be early canaries that signal broader changes—especially in how firms staff junior work.
4.3 Hiring competition and early talent signals (NACE + Handshake reporting)
NACE’s Job Outlook 2026 (published Nov 2025) reports:
- employers project a 1.6% increase in hiring for the Class of 2026 vs Class of 2025 Default
- 13.3% of jobs now require AI skills Default
- 10.5% of entry-level job posts now require AI skills Default
Separate reporting on Handshake data indicates:
- entry-level job postings on the platform were down over 16% year-over-year, and
- applications per job were up 26%, signaling increased competition. Business Insider
Interpretation for 2026:
Even if total hiring is “flat-ish,” requirements are shifting. The entry-level market becomes a skills filter environment: candidates who can demonstrate AI-enabled productivity gain an advantage.
5) Which jobs are most at risk of being “replaced” in 2026 (and why)
In 2026, the most realistic “replacement risk” is concentrated where tasks are:
- high volume (lots of similar cases),
- text/code centered,
- standardizable with clear success criteria,
- low consequence if wrong or easy to QC,
- and where AI can be inserted without a long physical deployment cycle.
Below are job clusters with elevated 2026 risk, grounded in exposure definitions used by OECD, PwC, and the observed entry-level effects in the Stanford paper.
5.1 Clerical, administrative, and routine office work
Why: heavy on documentation, form processing, scheduling, summarization, and standardized communications—excellent targets for GenAI + workflow automation.
- OECD defines AI exposure by task speedups using LLMs; its examples of “highly exposed” occupations include creative writers, interpreters, and programmers, and “exposed” includes roles like financial examiners (this task-based framework captures why routine office tasks are vulnerable). OECD
- PwC explicitly includes data entry workers as AI-exposed and frames “automatable jobs” as those where AI can autonomously complete many tasks. PwC
What “replacement” looks like in 2026:
- fewer coordinators per manager,
- consolidation of roles (one person supported by AI handles what 1.3–2.0 people used to do),
- more temp/contract work for exceptions and edge cases.
5.2 Customer support / contact-center work (especially Tier 1)
Why: high-volume repeated queries, scriptable procedures, huge case histories.
Evidence signals:
- PwC lists customer service as an example of “automatable jobs.” PwC
- Stanford’s paper highlights disproportionate entry-level effects in AI-exposed occupations and finds concentration where AI is more likely to automate. Stanford Digital Economy Lab
What “replacement” looks like in 2026:
- AI handles “Tier 1” inquiries; humans handle escalations,
- fewer junior agents; higher expectations for empathy, judgment, and complex issue resolution.
5.3 Junior software / routine coding tasks (not all software roles)
Why: AI can generate boilerplate, tests, documentation, and suggest fixes—reducing the need for junior “throughput” roles.
Signals:
- OECD’s “highly exposed” examples include programmers. OECD
- Stanford’s paper finds entry-level employment declines concentrated in AI-exposed occupations such as software development and customer service. Stanford Digital Economy Lab
What “replacement” looks like in 2026:
- fewer junior hires per senior engineer,
- more evaluation on system design, debugging, security, and product judgment (human complementarity).
5.4 Marketing content production and basic communications work
Why: fast generation of drafts, variants, localization, A/B content; easier than strategy and brand positioning.
What “replacement” looks like in 2026:
- smaller content teams producing higher volume,
- demand shifts toward prompt workflows, brand governance, and performance analytics.
(While our sources here focus more on exposure frameworks and entry-level effects than marketing-specific counts, the mechanism is consistent with the same task structure driving exposure definitions in OECD and ILO work.) OECD+1
6) Which jobs are least likely to be replaced in 2026 (and why)
Jobs are more resilient when they require:
- physical presence in unpredictable environments,
- hands-on dexterity, caregiving, or complex human interaction,
- high-stakes accountability (legal/medical responsibility),
- relationship building, leadership, or context-heavy judgment.
Even in these roles, AI is still likely to augment rather than replace:
- drafting notes,
- summarizing cases,
- scheduling,
- decision support,
- training and coaching.
This aligns with:
- McKinsey’s emphasis that automation potential is not a job-loss forecast and adoption takes time. McKinsey & Company
- PwC’s finding that job numbers and wages are still growing in many AI-exposed occupations, suggesting transformation rather than wholesale elimination. PwC
7) Productivity, wages, and why “replacement” is not the only story
7.1 AI is correlated with higher productivity growth in exposed industries (PwC)
PwC’s job-ad + company financial analysis finds:
- industries most able to use AI have 3× higher growth in revenue per employee PwC
- productivity growth (revenue per employee, 2018–2024) rises from 8.5% in the least-exposed quartile to 27.0% in the most-exposed quartile PwC
PwC explicitly notes it cannot prove causation with certainty, but the pattern accelerates after 2022. PwC
7.2 AI skills command a major wage premium (PwC)
PwC reports a 56% wage premium for workers with AI skills (up from 25% the year prior). PwC
7.3 Skills are changing faster than training systems can respond (OECD)
OECD finds:
- one in three job vacancies has high AI exposure OECD
- only 0.3%–5.5% of analyzed training courses deliver AI content in four countries OECD
Implication for 2026:
The bottleneck is likely to be human adaptation—training capacity, organizational redesign, and trust/governance—not purely the technology.
8) 2026 outlook: three evidence-based scenarios
Because 2026 outcomes depend heavily on adoption speed, workflow redesign, and policy constraints, a scenario view is the most honest way to forecast:
Scenario A: “Selective scale” (most likely baseline for 2026)
- Many firms deploy AI broadly, but only some redesign workflows end-to-end.
- Entry-level hiring continues to tighten in the most exposed occupations (consistent with Stanford’s findings). Stanford Digital Economy Lab
- AI skills become a standard requirement for a growing share of jobs (consistent with NACE’s 10.5% of entry-level posts requiring AI skills and 13.3% overall). Default
Scenario B: “Agent acceleration” (faster replacement of junior tasks)
- Agentic systems move from experiments to scaled deployments faster than expected (McKinsey: 23% scaling, 39% experimenting). McKinsey & Company
- Larger reductions in junior headcount demand in software support, back office, and Tier 1 support.
- Productivity rises, but workforce transitions become more painful without training scale-up.
Scenario C: “Trust & regulation drag” (slower replacement, more augmentation)
- Adoption slows due to risk, compliance needs, and model governance constraints.
- AI used primarily as copilots with stronger human oversight; fewer full workflow automations.
- Replacement pressures are weaker, but skill requirements still shift.
9) Leading indicators to watch in 2026
If you want to measure “AI replacing jobs” in real time during 2026, the most reliable early indicators are:
- Entry-level hiring and internship volumes in AI-exposed functions (software, support, clerical). Stanford Digital Economy Lab+1
- Share of job postings requiring AI skills, especially for junior roles. Default
- Within-occupation employment changes by AI exposure (task-based measures like those used by Penn Wharton and OECD). Penn Wharton Budget Model+1
- Agentic AI scaling rates (not just pilot counts). McKinsey & Company+1
- Training supply expansion (AI literacy and job-specific enablement), given OECD’s evidence of low current course coverage. OECD
10) Recommendations for 2026
For employers (practical, near-term)
- Redesign workflows, not just tasks. The biggest productivity gains come when AI is integrated end-to-end, not sprinkled in. McKinsey & Company+1
- Protect the entry-level pipeline. If AI removes “starter tasks,” create structured apprenticeships where juniors learn judgment, domain context, and system oversight. (Stanford’s evidence suggests entry-level workers are the first to feel impacts.) Stanford Digital Economy Lab
- Treat AI skills as a baseline competency in hiring, but validate with work samples. (NACE shows AI skills requirements are already measurable in postings.) Default
For workers and students
- Build AI literacy + domain depth (not just “prompting”).
- Document measurable outcomes: time saved, quality improvements, throughput gains. (PwC’s wage premium implies markets reward demonstrable AI capability.) PwC
- Prioritize skills that complement automation: stakeholder communication, problem framing, judgment, and accountability.
For policymakers and education systems
- Scale general AI literacy programs, not only advanced AI engineering tracks. OECD finds current training tends to skew advanced while broad literacy is needed. OECD
- Use labor-market data to target transition support at the most exposed cohorts (especially early-career). Stanford Digital Economy Lab+1
- Encourage transparent reporting on AI-driven workforce changes to reduce rumor-driven panic and improve policy accuracy.
Key quantitative indicators at a glance
| Indicator | Data point | What it implies for 2026 | Source |
|—|—:|—|
| Early‑career employment in most AI-exposed jobs | 13% relative decline (ages 22–25) | Entry-level hiring/retention risk in exposed roles | Stanford Digital Economy Lab Stanford Digital Economy Lab |
| U.S. jobs potentially exposed (≥50% tasks automatable) | ~42% | Broad task transformation; not equal to job loss | Penn Wharton Penn Wharton Budget Model |
| U.S. jobs completely exposed | ~1% | Full job automation exists but is limited | Penn Wharton Penn Wharton Budget Model |
| Global workers with some GenAI exposure | ~1 in 4 | Global reach; varies by income, occupation, gender | ILO International Labour Organization |
| Global workers in highest exposure category | ~3.3% | Highest “replacement risk” slice is smaller | ILO International Labour Organization |
| Urban vs non-urban GenAI exposure (OECD) | 1 in 3 vs 1 in 5 | Cities will see earlier/faster churn | OECD OECD |
| Org AI use (at least one function) | 88% | AI is mainstream; scaling still the challenge | McKinsey McKinsey & Company |
| Scaling agentic AI | 23% scaling; 39% experimenting | Automation may accelerate beyond single tasks | McKinsey McKinsey & Company |
| Wage premium for AI skills | 56% | Strong reward for AI-capable workers | PwC PwC |
| Skill change rate in AI-exposed jobs | 66% faster | Continuous reskilling becomes mandatory | PwC PwC |
| Vacancies with high AI exposure | 1 in 3 | Job requirements shifting rapidly | OECD OECD |
| Training courses with AI content | 0.3%–5.5% | Training supply lag will constrain transitions | OECD OECD |
| Entry-level job posts requiring AI skills | 10.5% | AI literacy is becoming entry-level baseline | NACE Default |
References used (selected)
- Stanford Digital Economy Lab: Canaries in the Coal Mine? (Nov 2025 update). Stanford Digital Economy Lab
- Penn Wharton Budget Model (Sep 2025): exposure, adoption, employment patterns. Penn Wharton Budget Model
- ILO (May 2025): refined global index of occupational exposure. International Labour Organization
- OECD (Apr 2025): AI exposure in vacancies + training supply metrics. OECD
- OECD (Apr 2025): urban vs non-urban exposure briefing. OECD
- PwC (Jun 2025): Global AI Jobs Barometer 2025 (wages, skills, productivity). PwC
- McKinsey (Nov 2025): The State of AI 2025 (adoption, scaling, agents). McKinsey & Company
- Stanford HAI AI Index 2025 highlights (investment, adoption, cost trends). Stanford HAI
- NACE (Nov 2025): Job Outlook 2026 (entry-level AI skill requirements). Default
