Jevon’s paradox - Does it Apply?
What is Jevons Paradox?
William Stanley Jevons first observed in the 19th century that improvements in efficiency don’t always reduce consumption—in fact, they often increase it. This idea became known as Jevons paradox.
At its core:
When something becomes cheaper and more efficient to use, people tend to use more of it—not less.
Jevons noticed this with coal during the Industrial Revolution: more efficient steam engines didn’t reduce coal usage—they increased it, because they made coal-powered industry more economical and widespread.
Classic Applications of Jevons Paradox
Before we jump to AI, it’s worth grounding this in familiar examples:
Coal → Steam Engines → More Industry → More Coal
Fuel-efficient cars → Cheaper driving → More miles driven
Efficient computing → Lower cost per computation → Explosion of software and internet services
Cloud infrastructure → Cheap compute → Startup boom and SaaS proliferation
The pattern is consistent:
Efficiency → Lower cost → Increased demand → Expansion of the system
The AI Twist: A Different Kind of Efficiency
Now we’re seeing this play out again—but with a twist.
1. The Disruption Target Has Changed
Previous waves of Jevons-style expansion largely affected:
Physical labor (farmers → factory workers)
Manufacturing
Energy systems
Today’s disruption is hitting:
Lawyers
Accountants
Radiologists
Financial analysts
Software engineers
These are knowledge workers, not industrial labor.
2. The Output Has Changed: Physical goods→ Virtual Goods
Historically:
Efficiency produced more physical goods (steel, cars, textiles)
Now:
Efficiency produces virtual goods (code, legal reasoning, financial models, diagnoses)
And this matters.
Physical goods:
Are scarce
Have natural demand limits
Require logistics, materials, and labor
Virtual goods:
Are infinitely replicable
Have near-zero marginal cost
Can substitute directly for high-value human labor
The “X in a Box” Economy
AI is rapidly turning expertise into commodities:
Lawyer in a box
Radiologist in a box
Accountant in a box
Financial advisor in a box
Consider the economics:
TaskBeforeAfterDraft a will$3,000~$25 in tokensLegal research$300/hour<$10 equivalentFinancial modelingDays of workMinutes WHAT DOES THIS MEAN - do I just not understand bc I don’t speak Engineer?
The result:
Everyone has access to near-expert-level output
The knowledge moat collapses
Pricing power evaporates
And critically:
The same task is being done—not more tasks.
Does Jevons Paradox Still Apply?
This is where things get interesting.
The Traditional Expectation
If Jevons holds, we would expect:
Cheaper legal services → more legal services consumed
Cheaper medical diagnostics → more diagnostics
Cheaper financial advice → more planning
But reality may differ.
Why This Time Might Be Different
1. These Markets Are Not Purely Demand-Driven
People don’t typically want more:
Legal disputes
Medical procedures
Financial complexity
These are necessity markets, not consumption markets.
You don’t buy legal advice like you buy sneakers.
2. Efficiency Targets Labor, Not Production
In prior waves:
Efficiency increased output (more goods)
In AI:
Efficiency reduces labor required for the same output
Example:
You still need one will
One diagnosis
One financial plan
But now:
It takes 1/10th the labor
3. No Natural “More Demand” Flywheel
Jevons relies on:
Lower cost → more usage
But here:
There’s a natural ceiling on demand
You don’t:
Get 5 wills because they’re cheaper
Seek 10 diagnoses because they’re faster
4. Simultaneous Cross-Industry Disruption
This is key.
AI isn’t disrupting one sector:
It’s disrupting all knowledge sectors at once
Software → disrupted by AI
Finance → disrupted by AI
Healthcare → disrupted by AI
Legal → disrupted by AI
There’s no “fallback industry” absorbing displaced workers.
The Software Industry: First to Feel It
Software is both:
The producer of AI
The first victim of AI
We’re already seeing:
1 engineer doing the work of 2–3
Rapid prototyping via agents
“Good enough” code generated overnight
But also:
Fewer engineers needed per product
Thinner differentiation layers
The Rise of Foundation Models
The value is concentrating in:
OpenAI
Anthropic
Google DeepMind
Everything else becomes:
A thin wrapper over increasingly capable foundation models
Implications:
Your “AI product” is often just:
Prompting
Orchestration
UI (which may disappear into agents)
And even that is shrinking.
The Gold Rush Phase (Right Now)
We’re currently in:
Peak demand for AI implementation
Companies racing to adopt AI
Hiring engineers, AI specialists
Building agents, workflows, copilots
This looks like Jevons:
More efficiency → more activity
But this may be temporary.
The Convergence Point
Over time:
Foundation models absorb domain knowledge
Specialized models emerge:
Legal models
Medical models
Financial models
And they achieve:
80–90% of expert capability
At that point:
You don’t need 20 experts
You need 5
Where Jevons Breaks
Jevons assumes:
Efficiency creates new demand
But here:
Efficiency may simply eliminate labor
Not because:
We produce more goods
But because:
The same goods require fewer humans
A Different Kind of Future
This isn’t:
Farmers → factories → middle class
This is:
Experts → automation → fewer experts
And that raises a harder question:
If knowledge work compresses across all industries simultaneously… where does displaced labor go?
Final Thought: Are We Still in Jevons’ World?
Jevons Paradox taught us:
Efficiency fuels expansion
But AI may be teaching us:
Efficiency can also compress entire labor markets—without expanding demand
We may still see:
New industries
New roles
New forms of value
But the transition may not follow historical patterns.
And Maybe the Real Question Is…
Not:
“Will demand increase?”
But:
“What happens when expertise becomes a commodity?”
Because when everyone has access to:
Legal reasoning
Financial intelligence
Medical knowledge
At near-zero cost…
We’re no longer just making things more efficient.
