Evaluation
Standing section
How we measure machine ability, why the measurements decay, and what a serious evaluation would have to look like.
5 essays
Contents/Sections/Labour & Work
Section
Labour & Work
Abstract
What automation has historically done to occupations, who absorbed the cost, and why aggregate employment figures answer the wrong question.
The problem in one paragraph
The public argument about automation and jobs is conducted almost entirely at the level of national employment totals, which recover, and almost never at the level of specific cohorts and regions, which frequently do not. This is a distributional question being answered with aggregate statistics, and the mismatch is not accidental: the aggregate is easy to measure and flattering to quote.
What we look for in this section
Historical cases with actual data rather than analogies to the loom. Occupational entry rates rather than dismissal counts, because the dominant mechanism of occupational decline is a closed pipeline rather than a mass layoff. Task-level exposure studies and, crucially, whether anyone has checked that task exposure translated into job displacement.
We also try to be specific about what changes within a job that survives. The residual work after automation is not a smaller version of the old work; it is whatever resisted automation, and that is often the least rewarding part.
Recurring themes
The bank teller case and the elasticity condition that is usually dropped when it is quoted. Wage polarisation and whether the current wave points at the same part of the distribution as the last one. The long timescale of organisational adjustment, and the recurring finding that firms are the bottleneck rather than the technology.
Annotated bibliographyAll essays
- 02
The Distance Between a Demo and a Deployment
Eye for AI · 21 August 2026 · 16 min read
Filed here as well as under evaluation because the gap it describes is absorbed by people: the residual work of checking, correcting and covering for a system that mostly works.
A demonstration proves that a system can succeed. A deployment requires that it rarely fail, in a specific way, for a specific person, on a Tuesday. These are almost unrelated problems.
- 03
What Automation Actually Did to the Typing Pool
Eye for AI · 7 August 2026 · 19 min read
The historical case the section is built on. It supplies the mechanism — a closed hiring pipeline rather than a mass dismissal — that the other pieces keep returning to.
The clearest historical evidence on automation and employment is not about looms or robots. It is about clerical work, and what it shows is more unsettling than either side usually admits.
Other sections
Policy
Standing section
Regulation, procurement, compute thresholds and open weights — read as decisions with trade-offs rather than as a contest between good and bad actors.
2 essays