A few weeks ago I was catching up with an old friend. He has been a senior software engineer for longer than either of us cares to admit, and he is exactly the kind of person who thinks carefully before he speaks.
His company had rolled out AI tools across the organization, invested in licences, run internal workshops, sent the all-hands message about embracing the future.
Six months in, he told me: nothing has really changed.
The team feels faster in individual tasks, maybe. But delivery timelines are the same. Strategic output is the same. The decisions landing on his desk every morning are the same decisions, arriving at the same pace, with the same level of complexity. The tools were there. The transformation was not.
I told him he was in very large company, but i had to investigate on my own.
Earlier this year, the National Bureau of Economic Research published a study covering nearly 6,000 executives across the United States, the United Kingdom, Germany, and Australia. The finding was striking in its bluntness: over 80% of companies reported no discernible impact from AI on either productivity or employment, despite adoption rates continuing to climb.
PwC’s 2026 Global CEO Survey, drawing on responses from 4,454 business leaders across 95 countries, found that 56% saw neither increased revenue nor decreased costs from their AI investments. Only 12% reported AI had both grown revenues and reduced costs simultaneously.
These are not small samples. These are not fringe results. This is the mainstream experience of AI adoption right now, and it sits in almost complete contrast to the narrative dominating every conference, every earnings call, and every vendor pitch deck.
The gap between what is being promised and what is being experienced is one of the defining strategic problems of this moment. And the reason most organizations are getting it wrong is not what they think it is.
There is a historical analogy that I keep coming back to, and it comes from an unlikely source: the electrification of American manufacturing in the late nineteenth and early twentieth centuries.
When factories first gained access to electric motors, most of them simply replaced their steam-powered central shafts with electric ones and kept everything else exactly as it was. The layout did not change. The workflows did not change. The organizational logic did not change. Electricity was plugged into a steam-powered world.
For decades, the productivity gains were modest and economists were puzzled. The technology was transformative in theory. The results were underwhelming in practice. It was only when companies began to fundamentally reimagine their factory floors around the new capability, moving from centralized power to distributed motors at every workstation, redesigning layouts, reorganizing workflows, that the gains arrived at scale.
The Federal Reserve Bank of San Francisco made exactly this point in a recent analysis of AI adoption. Replacing the motor is not transformation. Redesigning the factory is.
Most organizations today are replacing the motor. They are taking existing workflows, dropping AI tools into them, and measuring whether the tools make those workflows faster. Sometimes they do. But the bottleneck does not disappear. It moves. Research across more than 10,000 developers found that AI coding assistants increased individual task completion significantly, but pulled review time up by 91%, simply shifting the constraint downstream rather than removing it.
Speed at one stage of a broken process is not transformation. It is acceleration toward the same wall.
McKinsey’s 2025 AI research identified something that deserves more attention than it received: organizations reporting significant financial returns from AI were twice as likely to have redesigned their end-to-end workflows before selecting their tools. Not after. Before.
The technology was not the differentiator. The organizational thinking that preceded the technology was.
This maps to something I have written about before in the context of digital transformation more broadly: the failure rate of these initiatives has stayed stubbornly around 70% for years, across industries and geographies, and the reason is almost never the tools. It is the assumption that tools create change, when in reality tools can only accelerate or reveal the organizational structures that already exist.
An AI assistant dropped into a culture that does not share information clearly will produce faster, more polished versions of the same miscommunication. An AI tool given to a team without a clear decision-making framework will generate more options faster, and paralyze the same people in the same ways.
The technology surfaces what was already true. It does not fix it.
Deloitte’s 2025 survey of nearly 1,900 executives found that 85% of organizations increased their AI investment in the past year, and 91% plan to increase it again, even as most report achieving satisfactory ROI only within two to four years at best.
Investment continues because the fear of falling behind is real, and because the long-term potential of the technology genuinely is significant. I am not arguing against that. What I am arguing is that more investment in tools, without a corresponding investment in organizational redesign, will produce more of the same results.
The leaders who will look back on this period well are not the ones who moved fastest to deploy. They are the ones who asked the harder question first: what would we need to change about how we actually work for this technology to make a difference?
That question is slower. It is less visible in quarterly reporting. It requires the kind of organizational honesty that is uncomfortable in any room where the pressure to show AI progress is high.
It is also, based on everything the data is showing right now, the only question that leads somewhere real.
If this gave you a useful frame, pass it along to someone in the middle of an AI rollout who is starting to wonder why the results are not matching the expectations. They are probably asking the right questions already.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.