I contributed some technical content and opinions to an article titled “The AI Productivity Paradox” on the September 2026 edition of the Pure AI web site. See https://pureai.com/articles/2026/09/01/the-ai-productivity-paradox.aspx.

The article starts with the irresistible proposition of AI: give people AI, and they will get more done. A salesperson can draft an email in seconds, a programmer can generate code on demand. If every employee has a tireless digital assistant, it seems logical to conclude that productivity will rise.

However the article points out:

* AI may be increasing output without increasing meaningful productivity.
* AI has introduced a new “verification tax.”
* The real measure of AI productivity should be value, not volume.

Starting in the 1980s, putting word processing on every employee’s desk gave individuals the ability to create and revise documents themselves. Written communication became cheaper, easier, and faster, so we simply produced more of it: more memos, more reports, more proposals, more revisions, and later more email.

Tools such as VisiCalc (1979), Lotus 1-2-3 (1982), and Excel (1985) made sophisticated financial calculations accessible to virtually everyone. But businesses built bigger models, ran more scenarios, and came to expect analysis that previously would have been too expensive or time-consuming to produce.

AI lowers the cost of producing intellectual work, and when the cost falls, demand tends to rise. The productivity gain may therefore be real, but the amount of work required by an individual employee remains unchanged — or, more likely, increases.

I contributed some opinions:

McCaffrey observed, “Businesses tend to measure productivity by easy-to-count outputs: documents produced, lines of code written, calls handled, presentations created, or tasks completed.”

“But the history of technology suggests that output counting is a misleading metric when the technology itself makes output cheap. A factory that produces twice as many widgets is more productive. But a knowledge worker who produces twice as many documents may simply be generating twice as much organizational garbage.”

McCaffrey added, “The real test for AI shouldn’t be ‘How much more can we produce?’ It should be ‘What valuable things can we accomplish that we couldn’t accomplish before?’

Parrando’s Paradox is of of the most astonishing things I’ve ever encountered. Briefly, you have two completely different games that lose slowly but surely. However, by playing the two games in a certain sequence, you will actually win over time. Remarkable.

However, if you alternate by playing in the order C, C, S, S, C, C, S, S, and so on, incredibly, you will slowly win over time! Two losing games combine to give a winning game — absolutely astonishing.