AI Task Horizons Rose as Software Hiring Stayed Weak
METR 50%-success task horizon and Indeed software-development postings · model observations positioned on linear calendar time. Measured AI task horizons rose rapidly while software job postings remained below their February 2020 baseline. It places rapid capability gains beside weak software-labor demand without treating coincidence as causation.
What does it show?
Measured AI task horizons rose rapidly while software job postings remained below their February 2020 baseline.
Methodology
For every METR frontier-model release observation that overlaps Indeed's U.S. Software Development postings series, attach the latest weekly postings observation in the same or preceding month. The task horizon is METR's fitted 50%-success duration and belongs on a logarithmic scale. Indeed postings are indexed to February 1, 2020. Model releases are irregular, so this is not an evenly spaced time series. Both measures are affected by many forces; the display is a timing comparison, not a causal test of displacement. What this isn't: The coincident divergence is striking but is not evidence that AI caused the hiring decline.