Are We Paying for AI Twice?
For the last year and a half, I've been using AI tools, and I've seen the productivity gains—in engineering teams' work, faster documentation, sharper decision support, and even my own personal projects. But Satya Nadella raised something last week that I haven't stopped thinking about, and it's a point most leadership teams haven't caught up to yet.
He calls it the Reverse Information Paradox—the idea that we don't just pay AI vendors in subscription fees. We pay again, in something arguably more valuable: our institutional knowledge. Every prompt, every correction, every workflow we run through these tools is quietly teaching them how our business works.
Nadella built the idea on a much older one. Economist Kenneth Arrow described a classic problem in the economics of information: a seller who owns valuable information cannot prove its worth to a buyer without revealing it, and once revealed, the buyer has effectively received it for free. Nadella flips that dynamic for the AI era. With most enterprise AI tools, it is no longer the seller who is exposed—it is the buyer. To get real value from a model, an organisation has to feed it real information: how its processes work, how its best people make decisions, what its mistakes look like and how they get corrected.
That is the twice in paying twice. Once in the subscription invoice. Again in the institutional knowledge that gradually becomes part of how the model behaves—knowledge the provider retains long after the contract ends. Unlike a subscription fee, this second cost is invisible on any invoice. It shows up nowhere in a budget review, and because it is cumulative and gradual, most organisations will not notice it is happening until a competitor, or the AI vendor itself, is offering capabilities that look suspiciously close to home.
I have spent over two decades in the automotive industry, watching AI move from novelty to necessity inside engineering organisations, and the productivity case is no longer in question. Documentation that once lagged behind development by weeks now gets drafted in near real time. Decision support has genuinely improved. But every one of those gains runs on a feed of real information—codebases, architecture decisions, customer issue patterns, and the corrections engineers make when a suggestion misses the mark, which are often the most valuable signal of all because they encode exactly the tacit expertise that is hardest to document any other way.
It is worth naming specifically what is at stake. Proprietary engineering methods that took years of trial and error to develop. Customer insights drawn from support tickets and usage patterns. Internal processes—the shortcuts experienced people take that never made it into a formal document. The sum of all of this is, in effect, what makes an organisation different from the one down the street, and the Reverse Information Paradox says that difference is quietly leaking out, prompt by prompt, not through any breach but through ordinary use.
None of this is a reason to slow down AI adoption. It is a reason to ask better questions at the leadership level: which AI tools are actually approved for use with company data, and what data employees are allowed to share with them. Whether our most sensitive workflows deserve private or self-hosted AI instead of a public model. Whether responsible AI use means anything specific enough for an employee to apply, rather than a policy line nobody reads twice. Most organisations I have spoken with are still answering the first of these questions, and have not gotten to the rest.
AI is going to be a strategic partner in every industry. But the organisations that succeed will not necessarily be those using the most AI. They will be the ones balancing innovation with thoughtful governance, treating the knowledge they generate through AI use as an asset worth protecting, not an afterthought to the productivity conversation.
This isn't a productivity conversation anymore. It's a leadership conversation.
This reflection was inspired by Satya Nadella's comments on the "Reverse Information Paradox," as reported by NDTV: https://www.ndtv.com/artificial-intelligence/microsoft-ceo-satya-nadella-highlights-the-hidden-cost-of-ai-usage-for-companies-11769273