How this works
JobCooked is a toy built on real research. It shows three possible futures for 2030 — not a prediction of what will happen. Nobody knows which future we're in yet; Anthropic's researchers find the three paths look nearly identical until 2027.
1. The three futures
From Anthropic's Economic Scenarios for Transformative AI (September 2026), Table 3. All numbers are 2030 versus a world without AI.
| Scenario | GDP | Knowledge-job pay | Other pay | Knowledge-job unemployment |
|---|---|---|---|---|
| Slow roll | +1.6% | +0.4% | +1.1% | 2.9% |
| Speeding up | +8.3% | -0.3% | +5.9% | 4.5% |
| Full send | +32.4% | -11.5% | +33.6% | 17.9% |
2. Your job
Anthropic's model has two groups: knowledge jobs (management, business, computer, engineering, science, legal, education, media, healthcare practitioners, sales, office) and everything else. We place your job with O*NETjob titles, then scale the group's numbers by how exposed your specific job is to AI — using the "GPTs are GPTs" exposure scores and the Anthropic Economic Index (how much real AI use of your tasks automates the work versus assists a person). Jobs where AI mostly automates get hit harder, matching Stanford's 2026 "Canaries" findings. Pay is the May 2025 BLS OEWS median for your job in your state, grown at the 3.5% wage trend to 2030.
3. 2033 and 2036 (our extrapolation)
Anthropic's model stops at 2030. For 2033 and 2036 we keep each scenario's 2030 gaps growing along its adoption curve — the slow world keeps catching up (gaps ×2.2 by 2033, ×3.8 by 2036), the fast world less (×1.7, ×2.3), the extreme world is near saturation (×1.3, ×1.45). Unemployment gaps grow more slowly (square root), since people retrain and switch fields. Robots keep improving too: by 2036, up to 35% of physical-job hours in the extreme case. Treat these years as "what if the trend keeps going", not research.
4. Robots (our addition)
Anthropic's model assumes robots don't improve — physical work is off-limits. We don't think that's safe to assume. With 🤖 on, a share of physical-job hours gets automated by 2030 (Slow roll 0.5%, Speeding up 3.0%, Full send 10.0%), anchored to McKinsey's estimate that robots could technically do 13% of US work hours. Structured physical work (factories, warehouses, driving) is more exposed than messy hands-on work. For knowledge jobs, robots close the escape hatch: in Anthropic's model displaced office workers find work in physical jobs, so with robots on, their unemployment stays higher. Pay losses follow Acemoglu & Restrepo: about 1.2× the job loss.
5. Prices (our projection)
Today's home values (Zillow), rent (Census/Zillow), gas and electricity (EIA) and groceries (BLS) for your state, grown with ~2.2% inflation (Fed/CBO), homes ~2%/yr (Fannie Mae, Zillow) and power ~1.7%/yr (EIA) — then nudged by each scenario. Things made by physical labor get pricier as those workers' pay rises; housing tracks the bigger economy; electricity rises with data-center demand (Carnegie Mellon/NC State: +6–29% nationally by 2030). Rent is scaled from Census to today's asking rents (Zillow). These are rough, stated assumptions, not forecasts.
6. The score
Pay cuts, vanished jobs, extra unemployment and the share of your tasks automated add up to "pain", mapped onto 0–100. Raises soften it. It's a vibe with math behind it.
Credits
This site includes information from O*NET 31.0 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under the CC BY 4.0 license. O*NET® is a trademark of USDOL/ETA. JobCooked has modified all or some of this information; USDOL/ETA has not approved, endorsed, or tested these modifications. JobCooked is not affiliated with Anthropic.