NEW YORK — What happened
A multi-institute study published this month tracking over 50,000 knowledge workers across tech, finance, and legal sectors revealed a startling disconnect: internal software metrics show a 40% surge in individual output—measured by commits, documents edited, and messages sent—while corporate revenue per employee and product release velocity have remained entirely flat.
Why it matters
For years, productivity software vendors have pitched generative AI and automation suites as direct multipliers of human output. If workers can generate twice as many drafts or code snippets, conventional wisdom assumed companies would produce twice as much value. This study demonstrates that modern software has simply accelerated the generation of intermediate artifacts, creating a massive wave of internal noise and review overhead rather than accelerating final deliverables.
Deep dive
The phenomenon, sometimes called 'digital inflation,' occurs because modern tools make writing and generating content so frictionless that reviewing and verifying that content becomes the new bottleneck. Software developers, for example, can generate entire modules in seconds using coding assistants, but senior engineers now spend significantly more time reviewing, refactoring, and hunting down subtle logic bugs in AI-generated code. The net result is zero acceleration in shipping features, accompanied by a rise in cognitive fatigue.
Report check (claims vs what is verified vs still rumor)
Software toolmakers cite internal surveys claiming a 55% boost in day-to-day coding speed. Independent time-motion studies verify that typing and drafting speeds are indeed faster, but total cycle time from feature conception to production deployment shows negligible improvement. Rumors that corporations are beginning to penalize employees for excessive, unreviewed AI code generation are currently unsubstantiated.
Open questions
How will human resources and management structures evolve when traditional metrics of individual productivity—like lines of code or document volume—become completely decoupled from business impact?