The AI Productivity Paradox
The article discusses the AI Productivity Paradox, where companies increase output speed with AI but fail to improve outcomes. It attributes this to teams using AI to accelerate the old project model rather than the product model. Strong product teams use AI differently for discovery and delivery, achieving better results.
back to all articles
Product Operating Model July 23, 2026 Marty Cagan
share:
Share on Twitter
Share on LinkedIn
Share via Email
Copy the Link
The AI Productivity Paradox
We’ve been writing a lot lately about product teams that are clearly leveraging AI to deliver faster, yet their outcomes are not improving.
Today, this phenomenon, known as the “AI Productivity Paradox,” has been recognized by people from across the industry.
From the latest McKinsey Quarterly: “The business world is grappling with an AI paradox: Adoption of generative and agentic AI is growing, investment is accelerating, but sustained impact on performance is elusive.”
From the Atlassian’s State of Teams 2026 Report: “89% of executives say AI has increased the speed of work, but only 6% feel confident they can point to specific organization-wide AI ROI.”
Yet while it may be easy to agree that AI increases productivity but not necessarily results, there’s much less agreement on why this is.
For those of us that have been studying this problem of accelerating output without the corresponding improvement in outcomes since long before AI, this has not been a surprise.
And it isn’t really much of a paradox either.
We continue to see most people utilizing AI to simply speed up their old, project model way of working.
As AI product leader Hilary Gridley argues, “It’s never been faster to build, which means it’s never been easier to run 10 times faster in the wrong direction.”
As Chip Huyen, the author of the bestselling AI Engineering points out, “AI makes building easier, but the hardest part remains knowing what to build.”
The real problem with the project model was never that it was too slow (although it is often slow). The larger problem is that it’s designed to deliver output, rather than outcomes.
When generative AI emerged, especially as it pertained to building products, I was optimistic that it would serve as the great equalizer, and companies with the best engineers would no longer have such a strong advantage over the rest.
With the benefit of hindsight, it’s pretty clear now that those with the best engineers also often had the best product people, and that the true advantage was less their delivery skills, and more their culture, strategy and discovery skills.
So the result today is literally the opposite of what I had initially expected. Rather than closing the gap, the strong product companies are increasing the distance between themselves and the majority of the market.
The product model is what is enabling these companies to leverage AI for improved outcomes and not just output.
Recently I wrote about how strong product teams use AI very differently when they are building to learn (product discovery) versus building to earn (product delivery).
While so many are using AI to accelerate the creation of the artifacts of the old project model (e.g. business cases, roadmaps, PRD’s, code) the strong teams are using AI to accelerate the discovery of a solution that solves for both customers (value) and their own company (viability), and test those proposed solutions with users, customers, and the impacted stakeholders.
Once they have the evidence and confidence that they have a solution worth building, then they use AI to accelerate their building to earn – focusing on building a commercial quality product – a solution that is reliable, accurate, scalable, performant, and, more generally, something that their customers can depend on.
You might wonder why a team can’t just generate something quickly and launch it to customers and see what happens? They absolutely can, and that’s precisely what so many are doing today. The problem is the outcome. Hence the AI productivity paradox.
For so many company leaders, it doesn’t matter when I show them the data, or even point to their own results. They are deeply convinced that if they could just get their ideas built faster, the results will surely follow.
For many of these leaders, I expect we will simply have to wait until they can no longer deny the evidence that the issue is not time and cost of building; the real issue is that their ideas so often prove to be not worth building (they are simply not an effective solution to whatever problem they are trying to solve).
But for those who embrace the different purposes, tools and techniques of build to learn versus build to earn, there has never been a better time to be creating products powered by technology.