COBUS KOK

Companion to Cheap Composition · Draft 4, 7 November 2026 · Beta

The Personal Singularity

The singularity for an individual is the day the thing you're best at becomes cheap.

For each skill below, two dates. When a machine could do it, and when the people who do it felt it in their income. The gap between them interests me most. Law, liability and whether the work needs hands decide how long it is. How clever the model is doesn't come into it.

A date with a source is documented, and the source is linked in its row. Every other date is my estimate, with a range, whichever side of today it falls.

Where are you on it?

This names your row. The three answers set the dates.
This sets the capability year. The longer the task, the later a machine can do it.
10%
noneall of it
20%
noneall of it

The timeline

What a machine can do it felt in income the gap you
How sure solid: documented, source linked dashed, with a range: estimate

Tap or hover a row for its evidence, what gets cheap and what stays for now.

How the dates are made

Documented dates link to their evidence. For a capability date that is a benchmark result, a study or a deployment doing the work at scale. For a felt date it is a measured fall in pay, work or hiring. A date without that stays an estimate, even when it is already behind us.

Estimated capability years come from task length. METR measures how long a task AI can finish at 50% reliability. By early 2025 AI could finish one-hour tasks half the time, and that length kept doubling until the best models outgrew METR's tests in 2026. METR's January 2026 update put the doubling time at about four months since 2023 and about three months since 2024. This page uses three months, because at four months per doubling the rule would put 16-hour tasks about half a year after METR measured them. In May 2026 METR reported a model at 16 hours or more and warned that its tasks cannot measure reliably beyond that. It stopped updating those measurements in September. So the longer a unit of your work, the later the machine gets there. Hands-on physical work in unpredictable places adds up to eight years. That assumes robots trail language by about a decade.

The dates may come sooner. In November 2026 METR published a different kind of measure. After eleven weeks with confidential access at Anthropic, it estimated that the lab is likely improving its models on tasks with clear, checkable answers about twice as fast as it did from March 2024 to October 2025, and guessed that most of the speed-up comes from AI helping with the research. With all the labs' frontier models included rather than Anthropic's alone, the speed-up came out smaller, about 1.4 times. METR found no convincing sign that a shortage of new methods, computing power or data will hold back Anthropic's rapid progress over the next year, though it isn't confident of that. Anthropic removed from the public report most of the evidence METR found most informative. If the speed-up holds and spreads to other labs, the year a machine can do long work comes sooner than this page shows. Had task length been doubling 1.4 to 2 times as fast since March 2026, when METR measured 16 hours, work that takes a month would come about three to five months sooner than shown here. No test can check that yet. The year it shows in income would move by the same amount, because the gap between the two years depends on law, liability and hands-on work, not on how fast the models improve.

Estimated felt years add a lag. One year as a base, up to eight more for trust, and up to one more for hands-on work. Trust means law, licensing, liability and whether people need a specific accountable human. Those three numbers are my assumptions. Radiology is the calibration point. By 2019 a model could do one of its reading tasks, spotting lung cancer on low-dose CT screening scans, as well as or better than radiologists. Radiologists are still being hired. The calculator above shows every term for your answers.

The narrow models that read scans or drove cars arrived before general models did, which is why a few capability dates sit earlier than the task-length rule alone would give.

Sources

All rows as a table, with their evidence
SkillGroupA machine can do itFelt in incomeEvidence
TranslationComposition2023documented2023documentedCapability: Yan et al., 2024
Felt: Teutloff et al., Journal of Economic Behavior and Organization, 2025
TranscriptionComposition2022documented2023estimate, 2023 to 2026Capability: Radford et al., 2022
SEO and marketing copyComposition2022documented2023documentedCapability: Noy and Zhang, Science, 2023
Felt: Demirci, Hannane and Zhu, working paper 2024, published in Management Science 2025
Stock illustration and concept artComposition2022documented2023documentedCapability: Al Jazeera, September 2022
Felt: Hui, Reshef and Zhou, 2023
Tier-1 customer support (text)Composition2024documented2025documentedCapability: Klarna, February 2024
Felt: Fortune, September 2025
Voice narrationComposition2023documented2024documentedCapability: TechCrunch, January 2023
Felt: National Association of Voice Actors, 2024
Junior software developmentComposition2024documented2025documentedCapability: Anthropic, October 2024
Felt: Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, November 2025
Graphic production designComposition2023documented2023documentedCapability: TechCrunch, September 2023
Felt: Demirci, Hannane and Zhu, working paper 2024, published in Management Science 2025
Data analysis and reportingComposition2023documented2026estimate, 2025 to 2028Capability: Cheng, Li and Bing, EMNLP Findings 2023
Stock and background musicComposition2024documented2026estimate, 2025 to 2027Capability: Suno, March 2024
Research synthesis (analyst work)Composition2025documented2027estimate, 2026 to 2028Capability: OpenAI, February 2025
Video editing and motionComposition2025estimate, 2024 to 20262027estimate, 2026 to 2028None yet: both dates are estimates.
Senior software engineeringComposition2027estimate, 2026 to 20282029estimate, 2028 to 2031None yet: both dates are estimates.
Management and coordinationComposition2027estimate, 2026 to 20292030estimate, 2029 to 2033None yet: both dates are estimates.
Radiology (lung screening reads)Trust-gated2019documented2029estimate, 2027 to 2032Capability: Ardila et al., Nature Medicine, May 2019
Paralegal document reviewTrust-gated2023documented2026estimate, 2025 to 2028Capability: A&O Shearman, February 2023
Bookkeeping and tax preparationTrust-gated2025estimate, 2024 to 20272027estimate, 2026 to 2029None yet: both dates are estimates.
Contract draftingTrust-gated2025documented2028estimate, 2026 to 2030Capability: LegalBenchmarks.ai, September 2025
Teaching (instruction delivery)Trust-gated2024documented2030estimate, 2028 to 2034Capability: Kestin et al., Scientific Reports, 2025
Primary-care triage and diagnosisTrust-gated2024documented2031estimate, 2029 to 2035Capability: Tu et al., Nature, 2025
Therapy and counsellingTrust-gated2025documented2032estimate, 2029 to 2040Capability: Heinz et al., NEJM AI, March 2025
Driving (trucks, ride-hail)Embodied2020documented2029estimate, 2027 to 2032Capability: Waymo, October 2020
Warehouse pickingEmbodied2025documented2028estimate, 2027 to 2030Capability: Amazon, May 2025
SurgeryEmbodied2031estimate, 2028 to 20362036estimate, 2033 to 2040None yet: both dates are estimates.
Plumbing, electrical and the tradesEmbodied2032estimate, 2030 to 20382035estimate, 2033 to 2040None yet: both dates are estimates.
Nursing and hands-on careEmbodied2033estimate, 2030 to 20402037estimate, 2035 to 2040None yet: both dates are estimates.
Live performanceThe human is the productNot on this axis.
Professional sportThe human is the productNot on this axis.
ChildcareThe human is the productNot on this axis.

Made in September 2026 with Claude. Draft 4, November 2026: the model now uses METR's doubling time since 2024, about three months, instead of 4.3 months. The model behind "place me" is three answers, a doubling rate and a few stated assumptions. I wouldn't plan a life on it.