The Productivity Myth
A growing body of research argues that AI’s productivity gains are overstated, and it’s worth engaging directly rather than talking past it. Economists from the University of Chicago and the University of Copenhagen studied a group of 25,000 workers and found that although 90% of AI users believed the tools saved them time, the actual measured impact averaged just 2.8% of work hours, and a quarter of workers spent more time on identical tasks afterwards. Wage growth among even the heaviest AI users was, in the study’s own terms, near zero.
It’s not an isolated finding. MIT research puts the share of corporate generative AI investments producing zero measurable return at 95%; a figure serious enough to spark talk of a wider AI investment bubble. A separate line of evidence points the same way. Goldman Sachs’s Chief Economist, Jan Hatzius, put AI’s actual contribution to 2025 US GDP growth at “basically zero”, despite the scale of investment pouring into the sector.
Where the Skeptics Are Right
The critique holds up best at exactly the level most people experience AI: the individual chatbot session, the tool subscription, the borrowed prompt template. Workday research found that 40% of the time AI saves gets lost again to rework. BCG found productivity actually rises with three or fewer AI tools in someone’s workflow and drops sharply beyond four, with a third of workers reporting “AI brain fry” severe enough that they intend to quit. MIT and Stanford researchers have a name for the output this produces: “workslop” – polished-looking work that creates more rework than it saves, estimated to cost a 10,000-person organisation $8–9 million a year.
The common thread across all of it: none of these gains is owned by the person using the tool. The time saved (when it’s real) evaporates the moment the session ends. There’s no asset left behind, nothing that compounds, licenses, or grows in value the next time it’s used. That’s a legitimate limitation, and it’s the correct read of “AI productivity” as most people encounter it.
The Real Shift Isn’t Productivity; It’s Restructuring
Set next to that skepticism is a much larger and better-established body of research: not about whether a chatbot saves you ten minutes, but about how the shape of work itself is changing. The World Economic Forum’s Future of Jobs Report 2025 projects 170 million new roles created and 92 million displaced by 2030, a net gain of 78 million jobs, but with 22% of total jobs affected by the transition. Nearly 40% of the skills required on the job are expected to change in that window. 59 out of every 100 workers will need reskilling or upskilling; 11 out of 100 are unlikely to get it, leaving more than 120 million workers at medium-term risk.
That’s the real conversation, and it’s not a debate about whether AI makes any single task faster. It’s a debate about who is positioned to own and redeploy their own expertise as the underlying structure of work shifts underneath them, and who is left holding a skill set with a shrinking shelf life.
What Compounds and What Doesn’t
Put the two bodies of research side by side, and the pattern is clear: generic, tool-level AI productivity doesn’t compound, because nothing about it is owned by the person using it. The Fusion Economy is a different claim entirely, not that AI makes any one task faster, but that a person’s own structured, owned expertise, once captured and deployed as an asset, compounds the way a chatbot session never can.
| Generic AI Productivity | The Fusion Economy | |
| What it claims | “This tool saves you time” | Owned expertise compounds into new income |
| Who owns the gain | The platform, moment to moment | The person, permanently |
| Does it compound | No, near-zero wage impact after adoption | Yes, a structured knowledge asset, reused and multiplied |
| Evidence it’s working | 95% of corporate AI investments show zero return | Three Income Model: role + licensed expertise + compounding IP |
The Fusion Economy: A Different Productivity Claim
Bloor Research’s Fusion Economy describes what happens when a person stops renting productivity from a platform, session by session, and instead builds a durable asset out of what they actually know.
The Three Income Model names exactly how, through three streams:
- Your time: fixed and linear, roughly 2,000 hours a year. Every hour is sold once.
- Digital Me℠: your knowledge and IP operating as a sovereign digital asset, deployed 24/7 with no ceiling.
- FusionWork: human and digital labour working simultaneously, so income compounds instead of resetting to zero at the end of every session.
Neither of the last two streams depends on a chatbot answering faster. They depend on the underlying knowledge being captured, structured, and owned in the first place, exactly the piece the productivity-myth research shows generic AI use was never built to deliver.
Frequently Asked Questions (FAQs)
For the way most people use it – a chatbot session, a borrowed prompt – the research says yes, largely. A Danish study of 25,000 workers found near-zero measurable time savings and wage impact despite high adoption and high confidence.
Confidence and actual measured output diverge. 90% of users in the Danish study believed AI saved them time; the measured effect was 2.8% of work hours, and a quarter of users spent more time on the same tasks afterward.
That a person’s own captured, structured expertise, deployed as an owned asset rather than rented from a platform, compounds into additional income streams. That’s a claim about ownership, not about chatbot speed.
No, those numbers (170 million roles created, 92 million displaced, 40% of skills changing by 2030) describe a real structural shift. The productivity-myth research and the future-of-work research are about two different things: task-level speed versus workforce-level restructuring.
Sources
- World Economic Forum – Future of Jobs Report 2025 — job creation, displacement, and reskilling figures.
- Goldman Sachs — AI’s “basically zero” contribution to 2025 US GDP growth
- BCG’s “AI brain fry” findings, via Fortune — the 3-vs-4-tools productivity threshold.
- Stanford / BetterUp Labs -“Workslop” — the $8–9M-a-year cost estimate.
Print Article for this blog Download here