The compiler was supposed to end programming too

This explains how software automation has repeatedly been proclaimed the end of programming - starting with FORTRAN in 1954 - and argues that current AI coding agents repeat that pattern but are distinct for their speed and scale. It opens with a concrete 2031 vignette in which multiple autonomous agents run overnight, open pull requests with tests and recordings, and only stop to ask human judgment calls (for example, what to do about a refund to an expired card). The central claim is not that programming disappears, but that tedious, boilerplate, and implementation work will be handled by agents while humans move up to defining clear specs, writing tests, reviewing edge cases, and making policy decisions.

The piece backs this with concrete numbers: GitHub reports growth to ~180 million developers, Stack Overflow adoption rose from ~70% to 84% using AI tools from 2023-2025 while favorable sentiment fell (77%→60%) and distrust of AI accuracy outpaced trust in 2025. A METR randomized trial (16 developers, 246 issues) found AI-assisted runs took 19% longer despite developers feeling faster. METR’s broader benchmark shows the task-length agents can complete doubling roughly every six months (GPT-2: seconds; GPT-4: minutes; Claude variants: hours), implying agents will handle much longer tasks within years. Industry signals (Google and Anthropic usage numbers) support rapid uptake and a likely shift from typing code to supervising and specifying it.