Nonfiction · 2026
What forty years of research on automation says about your job now
Launch price: ₹299 in India · $5.99 internationally
The promise was that AI would save time. It can. But time saved and difficulty removed are not the same thing.
What happens to human work, judgment, and capability when machines handle more of the routine — that is the question this book examines.
This book draws on forty years of research in automation science, cognitive psychology, human oversight, and recent studies of AI at work to examine what happens to the human work around automated systems.
No hype. Evidence first.
Why supervising and verifying automated output can create a different kind of workload — and what the research shows about it.
What decades of automation research say about monitoring, vigilance, skill, and what happens to human involvement over time.
What recent field studies actually found about time, earnings, hours, and how task allocation shifts when AI is introduced.
Why fluent or confident AI output does not guarantee accuracy — and what that means for people who rely on it professionally.
Why the existence of a checking procedure does not automatically mean that errors will be caught.
What research does — and does not yet — tell us about human capability when machines handle more of the routine work.
Every figure in this book comes from a named, sourced study. Here are four.
In a large study covering workers across 11 occupations, approximately 85% of reported AI time savings were reallocated to other tasks. The study found no significant change in earnings or hours worked over the period examined.
Humlum & Vestergaard (2026), NBER Working Paper 33777In a field experiment covering 680,676 customer service conversations, AI-handled chats were 16.8% shorter. Customer ratings on those same chats fell by 0.412 points on a five-point scale.
Wang et al. (2026), arXiv:2605.14830 — Preprint, not peer-reviewedIn a prospective study of 5,140 medication doses, 92.5% of mandated double-checks were primed — one nurse had already shared information that could influence the checking nurse. Where checks were mandated, there was no statistically significant association with fewer errors.
Westbrook et al. (2021), BMJ Quality & Safety — Peer-reviewedAmong full-time AI users, workers whose primary mode was supervising AI output reported 14% higher mental effort, 12% higher mental fatigue, and 19% higher information overload than other engagement patterns. This is a survey comparison among AI users, not a causal experiment.
Bedard et al. (2026), Harvard Business Review — Survey studyThe first chapter — The Tired Part — introduces the central problem through a concrete scene and the first round of evidence. No email required.
Read Chapter 1 FreePDF ebook · Instant digital delivery
Secure checkout · Digital delivery after purchase
Ten chapters. Forty years of research. One question about your job.
Get the Ebook · ₹299 / $5.99