Aaron Levie wrote a piece this week on Jevons Paradox that’s worth reading.
The core idea: When coal became more efficient to use in the 1800s, demand for coal went up, not down. Counterintuitive, but it makes sense once you think about it. Efficiency unlocks use cases that weren’t economical before. The same pattern shows up with mainframes to minicomputers to PCs—each generation seeing 100X more units. Cloud computing making enterprise software accessible to any barbershop in the world.
Levie’s thesis is that AI agents are doing this to knowledge work. Not reducing it. Unlocking it.
I think he’s right, and I’m seeing it in my own work.
Over the past year I’ve built a bunch of things at PandaDoc using Claude Code. A sales automation pipeline that handles territory routing and contact filtering. A web app that generates personalized sales collateral. A competitive intelligence agent that runs every morning.
When I look at this list, the honest truth is: none of these projects would have existed a few years ago.
It’s not that they weren’t valuable. They were obviously valuable. But getting them built would have meant engineering resources, and engineering resources are finite, and there was always something more urgent in the product roadmap. The “I” in ROI was just too high.
Levie puts it this way: “The mistake that people make when thinking about ROI is making the ‘R’ the core variable, when the real point of leverage is bringing down the cost of ‘I.’”
That’s a useful frame.
The sales collateral thing is a good example. Before, creating a personalized one-pager took 30-60 minutes of formatting work. So reps either didn’t do it, or they sent generic materials, or they waited on marketing. The work existed in theory but not in practice because the cost was too high.
Now it takes a few minutes. And guess what happened? Reps started generating personalized collateral for everything. The demand for that work didn’t decrease. It increased by maybe 10X. Because it finally made sense to do it.
I think that’s Jevons Paradox in miniature. And I think it’s happening everywhere.
There’s another part of Levie’s piece worth highlighting. He mentions that AI agents “require management, oversight, and substantial context to get the full gains.”
I don’t think that’s a minor point. It might be the most important one.
The paradox only kicks in if you can actually use the tool. Coal efficiency led to more coal usage because there were engineers who knew how to build steam engines. Cloud computing led to software proliferation because there were developers who knew how to build SaaS products.
AI agents lowering the cost of knowledge work leads to more knowledge work… for people who can direct the agents effectively.
This isn’t everyone. At least not yet.
I’ve watched people try to use these tools and give up because they couldn’t get past the initial friction. The capability is theoretically available to everyone, but actually capturing it requires a skill that’s still unevenly distributed.
Maybe that changes as the tools get better. Maybe the skill gap closes over time. I’m not sure. But right now, the gap is real.
Levie ends on this idea that “today’s jobs become tomorrow’s tasks.” The work doesn’t disappear—it gets absorbed into a larger scope of work.
That matches what I’m experiencing. I’m not doing less. I’m doing more, and different. The automation I build frees up time, and then I find new things to automate, new problems to solve, new systems to build. The scope keeps expanding.
I’m not sure what the ceiling is. Maybe there isn’t one.
That’s both exciting and a little disorienting. But I think Jevons Paradox is the right frame for thinking about it. Efficiency doesn’t shrink the work. It expands what counts as possible.