Jevons Paradox for Knowledge Work

Summary

When a resource gets cheaper, people use more of it — not less. That counterintuitive pattern, Jevons paradox, is now showing up in knowledge work. As AI agents drive down the cost of coding, research, contract review, and other judgment-heavy tasks, the realistic outcome is expanded demand for that work, not mass displacement.

The shift

William Stanley Jevons observed in 1865 that more efficient steam engines increased total coal consumption. The same dynamic repeated with cars, storage, and bandwidth: efficiency unlocks use cases that were previously uneconomical.

SaaS democratized deterministic work — tasks with clear inputs and predictable outputs. AI agents are doing the same for non-deterministic work: judgment, creativity, ambiguity. When coding gets 5–10x cheaper, organizations don't write less software. They discover far more things worth building — internal tools, personalization, comprehensive testing, disposable apps for one-off needs.

The bottleneck shifts. Execution gets cheaper; specification, oversight, and context become the scarce inputs. Jobs don't vanish — they absorb today's tasks into a broader scope. What was a full role becomes a task inside a larger role.

Why it matters

Leaders planning workforce and technology strategy should expect more knowledge work, not a simple headcount reduction story. The organizations that benefit will be those that can direct cheaper intelligence effectively: clear requirements, domain expertise, eval discipline, and workflow integration.

The oversight constraint is real. AI can automate tasks within a job, but the full workflow still needs human judgment to produce value. Capacity for review, integration, and quality assurance becomes the binding limit — not raw model capability.

What to do

• Invest in specification skills — writing what should be built, not just how to build it
• Treat agent oversight as operational capacity, not an afterthought
• Scope pilots around workflows where cheaper execution unlocks net-new work, not just cost cutting
• Build eval and quality infrastructure as output volume grows
• Develop domain translators who can turn institutional knowledge into agent-ready context