The industrial revolution automated the work of the hands; this wave automates the work that can be described in words. Drafting, specifying, summarising, reporting and reviewing
— the substance of much white-collar work — are now within reach of automation, with one durable exception: the engineering loop of describing a problem precisely, testing several
candidate solutions, and taking responsibility for the one that ships. Drawing on a year of industrial practice rather than research, this talk presents concrete evidence from software
delivery at ITSF: presentations and documents generated as versioned code, delivery pipelines in which dedicated agents carry a ticket from backlog to review under blocking
quality gates, and two months of self-reported productivity gains. It argues that the differentiator is not the model but the harness, the codified skills and the adversarial review built around it. The talk then turns to consequences the audience may recognise: graduates
who no longer research before trusting an answer, experienced practitioners who can now run alone what once required a team, and the physical and environmental cost of this infrastructure — component prices, datacenter siting, power and water. No roadmap for
academia is offered. The observations are industrial, the conclusions are the audience's, and the discussion is expected to be contested.
