Preview • August 2026
The AI Productivity Paradox: Why Your Best Metrics Might Be Your Biggest Liars
Why rising output figures often signal a shrinking capacity to catch machine-assisted errors, and what the law already expects boards to do about it.
Tony Ridley, MSc, CSyP, FSyI, SRMCP. Originally published 28 August 2026.
Boards across every sector are celebrating rising productivity figures as generative AI tools spread through drafting, analysis and decision support. Output per head is climbing, adoption rates look impressive, and the business case for further investment appears to write itself. This article asks a harder question: what happens to an organisation's capacity to know when that output is wrong?
Decades of research into automation bias and the erosion of human vigilance under automated systems suggest that expertise is not a fixed asset that protects an organisation once it has been acquired. It depends on continued, active engagement with the work, and that engagement is precisely what many AI deployments quietly remove. The people best placed to catch a machine's error are often among the first roles cut in the search for efficiency.
Tony Ridley examines why conventional productivity metrics can no longer be trusted at face value in an AI-assisted enterprise, why seniority alone does not protect against this exposure, and why the relevant legal and governance duties are not waiting on future AI-specific legislation to take effect. He sets out the structural conditions that separate organisations building durable capability from those quietly accumulating liability, and the questions a board should be asking about its own validation capacity before an external party asks them first.
This is a member article. The full piece is available to members. Become a member to read it in full.