Here's a pattern I noticed: The people most convinced that AI can replace their workforce are the people furthest from the work.
When I talk to engineers and designers, they tell me AI helps them achieve specific goals. They get really excited, and everyone has great "so much better/faster/more" stories to tell. But they have a hard time pinpointing examples where it made their team vastly more productive. And they talk a lot about the necessity of human oversight, of hiring new talent. When I talk to CEOs, they tell me AI is already making their teams more productive and their teams are getting smaller.
How can that be?
It's information architecture.
Every layer of hierarchy filters what travels upward. Successes get reported. Friction gets absorbed. By the time a result reaches the executive floor, it has been stripped of the context that explains what it actually took: the four failed attempts, the manual cleanup, the edge cases someone quietly handled. The executive sees a capability. The team sees a tool that helps with some things under specific conditions.
Both views are rational given the information each level receives. That's the problem. An organization can be structurally configured to make its leadership systematically overestimate automation potential - and most are, because the org chart was never designed to transport friction upward. It was designed to transport summaries.
So when a CEO concludes that AI makes a fifth of the workforce redundant, ask what information that conclusion is based on. If the answer is gut feeling, it isn't based on anything real (well, maybe "shareholder value"). If the answer is Jira tickets, demos, dashboards, and vendor briefings, the conclusion isn't analysis. It's an artifact of the reporting structure.
The fix isn't smarter executives, although shorter information paths to the actual work might help. There is no fix. We simply don't know how widespread AI adoption across the entire organization will transform us long term. Yet. It's just too early. But we can start experiments and take wild guesses.
Let's just not present our wild guesses as facts.
Let's start experiments.
Thoughts? Find me on Bluesky.