When Efficiency Creates More Work: Jevons Paradox and the AI Revolution
What a 19th-century coal economist can teach us about the future of knowledge work
In 1865, English economist William Stanley Jevons made a counterintuitive observation about coal-powered steam engines. As the engines became more efficient, everyone expected coal demand to drop. Instead, it skyrocketed.
The paradox? Greater efficiency didn’t reduce consumption, it unlocked use cases that had never been economically viable before. Suddenly you could attach a small steam engine to a belt running along your factory ceiling, distributing power to machines across the floor. What was impossible at old efficiency levels became obvious at new ones.
I recently sat down with three people I’ve worked with for over a decade: Ian Painter, Adrian McKenzie, and Oli Deakin, to explore whether this same paradox applies to AI and knowledge work. We built and sold Snowflake Software together, growing it into an aviation analytics business that was eventually acquired. Now we’re all watching AI reshape the very industries we spent years building.
The conversation that followed was one of the most honest discussions I’ve had about what’s actually happening, beyond the hype and the fear.
The Identity Crisis Nobody’s Talking About
Adrian, now a director of technology, opened with something that stuck with me. He’s watching his engineering teams respond to AI tools in two very different ways.
Some engineers are struggling to adapt. Not because they can’t learn the tools, but because their identity is wound up in how they write code. The quality of their code is who they are. When AI changes how that code gets written, it feels like an attack on their sense of self.
Others see AI purely as a faster path to customer outcomes. They don’t care how the code gets produced, they care that it works and delivers value. For them, AI is just another accelerant.
This isn’t a skills gap. It’s an identity gap. And it’s happening across every knowledge profession right now.
The 80% Problem
Oli, who served as CTO alongside us, raised something that anyone who’s used Copilot or Claude for coding will recognise instantly.
