Big Tech’s Looming Capability Crisis

In 2016, Geoffrey Hinton argued that AI would soon replace radiologists, yet demand for radiologists has since surged. The reason, this article argues, is that AI reduced the cost of image analysis while increasing the value of complementary human capabilities: judgment, accountability, and apprenticeship. The same dynamic now applies to software engineering. Although AI can generate code cheaply and quickly, companies risk confusing code production with the broader work of engineering reliable, scalable systems. Unlike radiology, software lacks strong liability structures or professional oversight, making AI-related errors harder to detect and correct. As firms cut senior engineers and shrink apprenticeship pathways, they accumulate “capability debt” and “judgment debt” that may only become visible years later. To avoid dismantling the human expertise that gives AI-generated output value, leaders should implement software provenance tracking, require named human sign-off on AI-generated code, and establish accountability systems that make poor AI governance costly. The central lesson is that AI changes what becomes scarce: not output itself, but accountable human judgment.