RESEARCH / EXPERIMENTATION
FOR WHAT’S NEXT
Automatic translations. English is the original.
03 / TECHNOLOGY
Studying the gap between a capability becoming cheaper and people being able to use it productively.
A small business and a large firm can buy access to the same model and get very different value. Data, setup, review, and the ability to reorganise work shape the outcome. We are interested in which of those conditions can be changed—and who benefits when they are.
A 2025 study of 5,172 support agents found a 15% average gain in issues resolved per hour, with larger gains for less-experienced workers in that workflow.
Brynjolfsson, Li & Raymond / The Quarterly Journal of EconomicsMETR’s early-2025 developer study found a slowdown. Its 2026 follow-up also shows why selection and measurement make current effects difficult to estimate.
METROur hypothetical first-month example includes setup, service, and review. Cutting the service price by 90% reduces the combined cost by only 7.2%. Remove the one-off setup cost and the same price cut has a different effect.
Reduction in the combined first-month cost.
Without the one-off setup, the reduction is 36%.
Illustrative assumptions: setup effort valued at $120; service fee $12; review effort valued at $18. After the price cut, the service fee is $1.20. Time values are not necessarily cash payments. These figures are not observed user costs.
Use the same model and a synthetic supplier-catalogue task. Compare access alone with structured onboarding and workflow support. Separate the effect of a subsidy from the effect of helping someone complete the work.
A research plan informed by published evidence. Data collection has not begun.
An independent review of correctness and usefulness, including repairs to the output.
Setup, prompting, waiting, checking, and rework. Record time and cash separately.
Prior experience, assistance needed, and people who abandon the attempt.
Whether gains become less work, better quality, higher margins, or a lower price for the customer.
If a price subsidy produces the same gains as workflow support, cost may be the binding constraint. If support improves results but takes more effort than it saves, it needs a different delivery model. A high adoption rate alone would not settle either question.
Lower model prices are only part of the story. Data, workflow, review, and bargaining power determine who gets the gains.
READ THE ARTICLEWe want to build tools and support that make useful capability available beyond organisations already equipped to take advantage of it.
This page could not be opened. Your place is saved.