A data-grounded simulation · 6 regions · 6 task classes · 2026–2036

Who does the world's work — humans or AI?

Every past technology wave — farm→factory, computer→knowledge work — ended in net job gains. This model tests whether that pattern must repeat. It couples AI capability growth, industry adoption, robotics fleets, and new-work creation into one system, seeded with real 2026 data (ILO, IMF, IFR, METR, US Census, Morgan Stanley). Move the assumptions; the answer moves with them — that is the point.

What this is. An exploration instrument, not a forecast. Built in August 2026 over five sessions by directing AI tools against a written specification, then pressure-tested until the bugs stopped changing the answer; every equation runs in this page's own source. Pick a preset or move a dial, then read the verdict. What the first four versions got wrong →

Share of the world's work
Year: 2026
Done by humans90%
Done by AI & robots10%
202620282030203220342036
AI share of work · 2036
of total task-hours, this view
Net human jobs · 2026→36
demand for human work
Structural gap · 2036
labor force minus human-work demand
Peak churn year
most workers displaced in one year

Reading this scenario

The world's work, reallocated

Three bands: work done by humans, work done without human hours (substitution + tool-absorption inside surviving jobs), and work that has been ELIMINATED outright — intermediation that stops existing once agents talk to agents. Augmentation compresses headcount before substitution, which is why 2026 starts near 7%, not 1%.

Human workAI & robot workWork eliminated outright

Net gains — or a structural gap?

Demand for human workers (including brand-new occupations) vs. the labor force. When the amber line falls below the grey line, the gap is people the economy no longer needs at current wages — the honest measure of "net loss."

Demand for human workersLabor forceStructural gap

Your two axes: cognitive→physical, and how fast each falls

AI's share of work in each task class, by year. Top rows are pure knowledge work (falls to software alone); bottom rows are physical work (protected until robot fleets scale). Interpersonal work resists longest — people pay for people, and the Klarna reversal shows why: reliability thresholds are real, not just cost curves.

The churn ledger

Each year: workers displaced (red, down) vs. re-absorbed by demand growth + hired into new occupations (amber, up). The line is the net.

DisplacedRe-absorbed + new work

Where the new work appears

Cumulative new-occupation jobs by 2036, by task class — the model's answer to "where do the gains come from."

Six economies, six different decades

Change in demand for human workers, 2026→2036 (%), and AI's 2036 work share by region. India and the rest of the world are shielded by physical/agricultural work mixes and slow diffusion — and exposed exactly where they export cognition (IT services).

Methodology — so it isn't hand-waving

A task-based labor model (Acemoglu–Restrepo family): jobs are bundles of tasks; AI absorbs task-share as capability × adoption; three counterforces fight back — demand growth from cheaper output, new-occupation creation, and the physical bottleneck of robot fleets. Annual steps, 2026–2036, six regions × six task classes.

1 · Baseline: who does what today

Global employment ≈ 3.55 billion (ILO-modeled). Regions: US 168M · China 735M · EU+UK 250M · India 600M · Other advanced 155M · Rest of world 1,640M. Each region's work-hours are split across six task classes, calibrated to ILO sector shares (agriculture 26% of world employment) and IMF exposure findings (≈60% of advanced-economy jobs AI-exposed vs ≈40% in emerging markets):

  • Routine cognitive — clerical, support, bookkeeping
  • Non-routine cognitive — coding, analysis, law, finance, design
  • Creative & strategic — research, senior judgment, invention
  • Interpersonal & care — health, teaching, hospitality, relationship sales
  • Routine physical — assembly, warehousing, driving, field agriculture
  • Dexterous physical — trades, repair, construction, complex manual work

2 · Capability: stacked S-curves, not a straight exponential

Cognitive capability follows a logistic (S-curve) per task class whose speed comes from the doubling-time dial (METR: ~7-month doubling in autonomous task length 2019–25, ~4 months in 2024–25; frontier agents at ~half-day tasks by early 2026). A 0.35 translation factor discounts benchmark progress into real-task coverage — a doubling of task-length horizon does not double the share of jobs' tasks covered. Each class has a ceiling (routine cognitive 95% … interpersonal 50%, a human-preference floor). Capability is not one curve: Wave 2 (the arrival dial) models foundation models transferring to robots, re-accelerating physical and interpersonal capability as a second S-curve stacked on the first — the "saturate, then a new round" pattern of every prior technology. And the ceilings themselves are generation-indexed: preference floors are measured against today's alternative (a Gen-1 bot doing Gen-3 work), so as capability generations advance the ceilings drift upward (interpersonal 50%→72%, creative 65%→85%), holding only an irreducible core for genuinely relational work.

3 · Adoption: tipping points, not constant crawl

Adoption is a logistic per region×class — but its speed is coupled to capability: below a reliability threshold (≈35–55% of tasks, by class), adoption runs at pilot speed (~0.15× base — the slow phase we are in now); once capability crosses the threshold, speed jumps ~1.85× (the launch). For physical work, speed is additionally scaled by wage-vs-machine cost parity: robot labor at ~$10/hr all-in (2026, falling ~16%/yr) tips against a $36/hr US wage years before a $2.50/hr Indian wage — cheap human labor is its own protection. 2026 starting points anchor to US Census BTOS (18% of firms, 32% employment-weighted, 50–70% in large information/professional/finance firms) and Stanford AI Index 2026 (88% of surveyed orgs). Regional base speeds: China ≈ US > other advanced > EU > India > rest of world; India's IT-export sector adopts at client-country speed.

4 · Robotics: the physical bottleneck

Physical task classes cannot automate beyond the installed fleet, in worker-equivalents: industrial robots 4.66M units operating in 2024, 542k installed/yr, ~2 worker-equivalents each (multi-shift), growing 8%/yr; autonomous vehicles from ~100k (Waymo alone: 450k+ paid rides/week in 2026), growing ~60%/yr; humanoids from ~150k units in 2026 (Morgan Stanley: 50k China shipments in 2026 → 446k/yr by 2030, 13M in use by 2035; Tesla targeting 10k→50k→1M/yr capacity), ramping on a logistic manufacturing curve — near-exponential early at the fleet-growth dial's rate, saturating toward a 250M-unit carrying capacity (factories are S-curves too; for scale, world vehicle production is ~90M units/yr). Worker-equivalence is explicit, not 1:1 — utilization asymmetry means one machine ≠ one job: industrial robots 2.0 equivalents (multi-shift); autonomous vehicles 2.5 (a robotaxi runs ~20 hr/day vs an 8-hr driver shift and truckers are capped at 11 driving hours — today's math: Waymo's 500k weekly rides across ~3,000 vehicles ≈ 165 rides/vehicle/week vs ~100 for a full-time rideshare driver, ≈1.6 driver-workloads per vehicle and rising with utilization); warehouse AMRs 1.0 (Amazon alone crossed 1M robots against ~1.5M employees, with its Vulcan system handling 75% of item types); humanoids 0.8→1.6 as uptime improves. The AV fleet starts from ~10k truly driverless commercial units in 2026 (Waymo ~3,000, Apollo Go ~1,000, Pony.ai 1,400+, WeRide ~1,000) growing ~90%/yr on a logistic capped at 25M; AMRs from ~2.5M at 30%/yr capped at 40M. Fleet is allocated China 45% · US 18% · EU 12% · other advanced 10% · India 5% · RoW 10%.

5 · The three counterforces

(a) Demand elasticity: cheaper output → more demand → workers re-absorbed in-industry (why ATMs coincided with more bank tellers for 20 years). (b) New-task creation: new occupations = elasticity × displaced work, with a 2-year lag, allocated 40% non-routine cognitive · 25% interpersonal · 15% creative · 12% dexterous physical · 8% other. Crucially, new tasks join the automatable pool — they are made of tasks too, and AI eats them at their class's automation rate in later years. The optimists' accounting error is granting new work permanent immunity; this model does not. Two further mechanisms: the tool→worker transition — below full substitution, AI compresses headcount on remaining work (augmentation: 0.55 productivity factor on cognitive classes, 0.25–0.35 on physical/interpersonal), and substituted work retains only a supervision residual of 1 human per span-of-control dial digital workers, a span that widens as reliability grows. And a chain-completion kicker: when make + move + coordinate (routine physical, routine cognitive, non-routine cognitive) jointly cross ~45% automation in a region — the fully autonomous build→warehouse→ship→sell→deliver chain — coordination labor collapses by up to a further 35%, capturing the synergy of automation vectors compounding each other. (c) Background growth: GDP-driven demand growth per region (India 5.5%/yr … EU 1.2%/yr) and labor-force drift (China −0.3%/yr, India +1.2%/yr, RoW +1.6%/yr).

6b · Obviation: work that stops existing

Distinct from automation: when your agent talks to their agent, the interaction is not performed by AI — it ceases to exist. No chat queue, no escalation, no scheduling email. Switchboard operators weren't replaced by robot operators; direct dialing deleted the task. The model tags each class's intermediation share — work existing only because humans can't interface directly (routine cognitive 45% · non-routine cognitive 20% · interpersonal 15% · creative 5% · physical 2–5%) — and eliminates it as consumer personal-agent penetration (a logistic gated on high agent reliability, from ~3% in 2026) meets enterprise-side agent adoption, scaled by the agent↔agent obviation dial. Eliminated work leaves the pool entirely — it consumes no AI compute, no robots, and creates displacement with nothing to supervise. It appears as the third band in the top chart. The displaced value recycles through the same new-task engine as all displacement.

6 · What the model does NOT claim

It does not predict. It makes the debate quantitative: the "past repeats" case is literally the new-task-elasticity dial near 0.85; the pessimist case is the same dial near 0.3 plus fast capability. It also omits wages (displacement shows up partly as wage compression before job loss), policy response, and AI-driven scientific acceleration beyond what the new-task dial captures — all of which would matter in reality. The model's reliability gates cut both ways, and the evidence demands both: tool penetration is deeper than surveys suggest (Google reports ~75% of new code AI-generated; young software developers −20% since 2022), yet durable unsupervised substitution keeps failing below the threshold — Klarna publicly walked back its showcase AI customer-service replacement and rehired humans for judgment-heavy work after satisfaction dropped, and surveys find a majority of firms regretting AI-first staff cuts. That is exactly the tool→worker distinction: augmentation compresses now; substitution waits for reliability. Read the Klarna reversal through the model's gates: it is what mid-gate premature adoption looks like — a Gen-1 product doing Gen-3 work — and the gates are indexed to capability, not calendar, so when a Gen-5 product does Gen-3 work the same deployment succeeds and adoption rips. Deeper still, the honest epistemic limit: obviation of enumerable intermediation work is modelable; wholly new interaction patterns that delete work categories we cannot yet name are not — and symmetrically, neither are the new categories they create. The obviation and new-task dials bracket that uncertainty rather than pretending to resolve it. Early real-world signal worth carrying: entry-level (22–25) employment in the most AI-exposed US occupations fell ~13% relative since ChatGPT while older cohorts in the same jobs grew — displacement is arriving rung-by-rung, youngest first, exactly as a task model predicts.

Data anchors & sources

  1. ILO / FAO employment baselines — ~3.5B employed globally; agriculture 916M (26.1%) in 2023. FAO/ILO employment indicators (July 2025 update)
  2. IMF, "Gen-AI: AI and the Future of Work" (SDN 2024/001) — 40% of global employment AI-exposed; 60% in advanced economies (27% high-complementarity / 33% low), 40% EM, 26% low-income. imf.org
  3. Autor et al. (2024), via IMF SDN 2026/001 — 60% of 2018 US employment is in occupations that did not exist in 1940; the anchor for the new-task-elasticity dial. imf.org
  4. IFR World Robotics 2025 — 4.66M industrial robots in operation (+9% y/y), 542k installed in 2024; China 54% of installs and 5× US stock; densities: Korea 1,220 / Singapore 818 / Germany 449 per 10k mfg workers. ifr.org
  5. Morgan Stanley (2026) — China humanoid shipments ~50k in 2026 → 446k/yr by 2030; ~13M humanoids in use by 2035; $5T ecosystem by 2050. cnbc.com · morganstanley.com
  6. Goldman Sachs Research — humanoid TAM $38B by 2035, ~1.4M units/yr. goldmansachs.com
  7. Waymo / physical-AI tracking — 450k+ paid robotaxi rides/week across five US cities (2026); Tesla Optimus ~1k units deployed internally, 10k targeted 2026. mlq.ai (source page no longer available)
  8. METR, "Measuring AI Ability to Complete Long Tasks" + Time Horizon 1.1 (2026) — autonomous-task time horizon doubling ~every 7 months 2019–25, ~4 months in 2024–25; similar rates across science, math, robotics, computer-use benchmarks. metr.org
  9. US Census Bureau BTOS AI Supplement (CES-WP-26-25, 2026) — 18% of firms using AI (32% employment-weighted), 50–70% among large firms in information/professional services/finance. census.gov
  10. Stanford AI Index 2026 — 88% organizational adoption among surveyed firms; 53% of global population using genAI within 3 years; software-developer employment ages 22–25 down ~20% from 2024. summary
  11. Brynjolfsson, Chandar & Chen (Stanford/ADP, 2025–26) — "Canaries in the Coal Mine": 22–25-year-olds in most-exposed occupations −3.8%/yr and accelerating; ~13% relative decline since late 2022; older cohorts +6–9% in the same occupations. fortune.com
  12. Robotaxi market state (2026) — Waymo 500k weekly paid rides across 10–11 US cities on a ~3,000-vehicle fleet, pricing ~15% below Uber/Lyft in overlapping markets; Apollo Go 20M cumulative rides, ~1,000 vehicles, unit-profitable in Wuhan; Pony.ai 1,400+ and WeRide ~1,000 commercial vehicles. robotaxi statistics · fleet analysis
  13. Amazon robotics (2026) — warehouse fleet crossed 1M robots; Vulcan touch-sensing system handles ~75% of fulfillment item types. report
  14. AI-generated code (2025–26) — Google ~75% of new code AI-generated (Pichai, 2026); Microsoft 20–30%; Snap 65% with headcount adjustments; Anthropic 90%+. devops.com
  15. The reliability counterweight — Klarna, after claiming its AI did the work of 700 customer-service agents, publicly reversed and rehired humans for judgment-heavy disputes as satisfaction fell; majority of surveyed firms report regretting AI-first staff replacement. summary

What I got wrong, and what changed

Five versions in three weeks. Each one was pressure-tested against how technology has actually diffused before, and each test found something the previous version had gotten wrong. The log is here because a model whose corrections never move its conclusions is not being tested.

v1Aug 2026 · first build

Task-based engine (Acemoglu–Restrepo family), six regions × six task classes, five dials, four presets. Net human-work demand by 2036 ranged from +984M to −100M across presets.

What was wrong. Capability grew as a clean exponential while adoption had no historical shape — no slow phase, no tipping point, no saturation. And the humanoid fleet compounded with no factory constraint: the fast preset implied 4.4 billion humanoids by 2036, more than the number of people who work.

v2adoption gets an S-curve

Adoption speed gated by capability — pilot pace below a per-class reliability threshold, launch pace above it. Regional wage-versus-robot cost parity (a $36/hour wage tips years before a $2.50/hour one). A Wave 2 dial for embodied AI as a second S-curve. Humanoids on a logistic factory ramp capped at 250M units. New-occupation jobs join the automatable pool instead of receiving permanent immunity.

What was wrong. Still too rosy. It priced automation by cost rather than output per dollar — one robotaxi does the work of more than one driver. It treated AI as a tool that makes a human more efficient, when the endgame is AI doing the work with one human overseeing many. And its 2026 starting point (~1% of world work) sat below what robotaxis, AI warehouses, AI call centres and AI-written code already implied.

v3tool becomes worker

Two-regime labour: augmentation compresses headcount on remaining work; substitution leaves a supervision residual of one human per span of control (new dial). Explicit worker-equivalents per machine type — autonomous vehicle 2.5, industrial robot 2.0, warehouse robot 1.0, humanoid 0.8 rising to 1.6. A chain-completion kicker when make + move + coordinate jointly cross 45% in a region. 2026 seeds recalibrated to ~7% of world work.

What was missing. A whole category. When my agent talks to your agent, the customer-service call is not automated — it never happens. That work leaves the economy instead of moving to a machine, and nothing in the model could represent it.

v4work that stops existing

An obviation channel: each task class carries an intermediation share that is eliminated — not automated — as consumer personal agents meet enterprise agents (new dial, and a third band in the top chart). Generation-indexed ceilings: human-preference floors erode as later generations do earlier-generation work. Consumer agent penetration as its own gated logistic.

v4.1the bugs

A fourth review pass found two errors that had been quietly shaping every headline number. Phantom churn: the ~246M of baseline tool-absorption at year zero was counted as year-one displacement, which corrupted the churn ledger, the peak-churn card and — through the two-year lag — the new-task creation of 2028. Year zero now establishes state without churn. Step discontinuity: Wave 2 arrived as a step and produced a visible kink; it now phases in over two years.

What it changed. Fixing them moved the answers materially: steady-diffusion net human demand went from +244M to +135M, the structural gap from 335M to 444M, and peak churn moved from a spurious 2026 to 2033. Those are the numbers this page now shows.

v4.2Sep 2026 · publication

Preparing this page for the web, an automated check found that the saved v4.1 file had lost its two label tables in a late edit: the heat map, the new-work chart, the regional chart and the written verdict had been failing silently for a month while the headline cards kept rendering. Labels restored; a console-error check is now part of the publish step. Also added: every scenario encoded in the URL, a copy-link button, and this log.

Link copied