Jevons’ Paradox Won’t Save You From the Bread Line
AI that is "good enough" will annihilate lawyers’ jobs.
That threshold — "good enough" — is the key to dismantling the optimistic belief that lawyers can comfortably co-exist with AI. Perfection is unnecessary. AI only needs to produce work that is adequate: meeting the client's needs at an acceptable level of risk. Once AI clears that bar for a task, the work stops going to humans.
We are arguably already past that point. For a wide range of routine legal tasks, AI is not just good enough — it is better than most lawyers. And yet, perhaps to comfort themselves, many in the legal profession have invoked Jevons' Paradox as their salvation.
The argument goes: AI will make legal services cheaper, which will unlock massive unmet demand, which will create more work for lawyers, not less.
It sounds plausible. It is wrong.
What Jevons Actually Said — and Why Coal Is the Wrong Analogy
William Stanley Jevons observed in 1865 that more efficient steam engines didn't reduce coal consumption. Conventional theory at the time sugggested that improving efficiency should reduce consumption. Instead, cheaper energy meant more applications for energy, so total demand for coal went up, not down.
The argument applied to legal AI follows the same structure: AI-driven efficiency will lower the cost of legal work, which will expand demand for legal hours, which will create more work for lawyers.
Except that Jevons' Paradox is a price mechanism. It only works when price is what drives demand in the first place. Lower the price of coal, and factories burn more of it — because the decision to use coal is directly responsive to its cost. Cheap enough coal means more coal-powered machines, longer operating hours, new applications that weren't viable before. Price is the lever, and it has plenty to pull on.
Legal demand does not work this way. It is not driven by price. It is driven by external events that have nothing to do with cost — a dispute arises, a deal needs to close, a company incorporates, an employee needs to be terminated. These events determine whether legal services are needed. Their frequency is not meaningfully altered by what legal services cost.
Nobody incorporates their company twice because AI made it cheaper. Nobody gets into more commercial disputes because legal advice became more affordable. Nobody signs more contracts because contract review dropped in price. The underlying business events that generate legal need happen on their own schedule, for their own reasons — and cheaper legal services do not make those events happen more often.
This is why Jevons' price lever has nothing to pull on in legal services. Lowering the cost of coal reveals new uses for coal. Lowering the cost of legal work does not reveal new legal problems — it just changes who, or what, handles the ones that already exist. And when "good enough" AI handles them instead of a lawyer, the demand for legal labour doesn't expand. It disappears.
The "92% of Unmet Legal Need" Argument
One most commonly cited supporting evidence for Jevons' Paradox applying to legal is the statistic that roughly 92% of legal needs go unmet — implying a vast reservoir of latent demand that cheaper, AI-powered services will unlock.
Let's pull it apart.
Unmet legal need has two very different sources:
Price-sensitive demand: People and businesses who want legal help, could benefit from it, but can't justify the cost relative to the value at stake.
Structural non-demand: People who cannot pay for legal services at any realistic price point — where the problem exists, but the paying capacity does not.
For the first category, yes — if AI makes a contract review cost $20 instead of $500, some businesses that previously skipped it will now act on it. That is real. However, in this category, the "good enough" AI may mean those clients end up choosing to use AI tools directly, rather than cheaper lawyers enabled by AI. Prior to reducing costs, these clients were choosing to skip legal review because the cost wasn't worth it to them. When AI is good enough, they may choose to use the AI — embedded in whatever platform they're already using — and the lawyer never enters the picture.
For the second category — the low-income populations that make up the bulk of that 92% figure — these prospective clients are not waiting for lawyers to become more efficient. They cannot pay for professional services at any price point that sustains a legal career. When AI serves them, it will do so through self-serve tools and embedded consumer apps at near-zero cost. The need gets met, and no lawyer gets paid.
We have to distinguish between latent demand and effective demand:
Latent demand is a desire for something.
Effective demand is a desire backed by the willingness and ability to pay for it.
The "unmet legal need" argument for Jevons' Paradox conflates the two — and ignores that “good enough” AI potentially eliminate the lawyer from the transaction for both groups.
The ATM Story Everyone Misses
The ATM case study is another pillar of the Jevons' Paradox argument.
When ATMs arrived, teller employment rose. The machine didn't kill the teller job. Proof that automation creates more work, not less.
This story has two chapters. Everyone only tells the first one.
Chapter one: In the 1970s, ATMs arrived, banks used the cost savings to open more branches, and teller headcount actually increased — imagine a bank had 10 tellers at one branch, and then they opened another four branches with three tellers each. 15 tellers instead of 10. This looks like Jevons' Paradox, more or less.
Chapter two: In the 2000s, mobile and online banking arrived. People go weeks without touching cash or visiting a branch. Teller employment collapsed — not because technology got more efficient, but because the underlying transaction migrated to a completely different delivery mechanism. People stopped needing to visit banks.
Now, consider, is legal AI more like the ATM, or more like mobile banking?
The ATM made cash withdrawal cheaper and more accessible, but cash withdrawal remained the thing people needed. The underlying transaction persisted. Jevons' Paradox applied because demand for cash didn't disappear.
Mobile banking didn't make cash withdrawal cheaper. It made cash withdrawal unnecessary for most purposes. Demand for cash didn't expand — it evaporated — but the demand for money persisted.
In our opinion, legal looks far more like mobile banking than the ATM. AI renders the human labour optional — and then, quickly, redundant.
When Routine Legal Becomes a Feature, Not a Separate Service
One of the most underappreciated consequences of "good enough" AI is what happens when output quality crosses the threshold for routine tasks. It doesn't just get cheaper. Instead, the capability becomes embedded closer to the source that needs the work done:
Contract drafting becomes a feature of procurement software.
Employment agreements become a feature of HR onboarding platforms.
Compliance checking becomes a feature of payroll systems.
Vendor term review becomes a feature of spend management tools.
Consider AI translation. After machine translation got “good enough” for everyday use. It didn't become a cheaper professional service. It became a button in Chrome. The latent demand for translation did not cause people to hire a translator for routine correspondence anymore — because machine translation cleared the "good enough" bar and got absorbed into the tools people were already using. The same people who would never have paid for professional translation simply got their needs met for free.
Once AI output is “good enough”, the service stops being separately rendered and becomes a feature of a workflow — invisible, automatic, and no longer a source of human employment.
Legal is on the same trajectory. The relevant question is not "will people demand more legal services as costs fall?" It is "will people demand legal labour at all, once legal outcomes are embedded in the software they're already using?"
Even Partial Automation Breaks the Headcount Maths
Let's be generous. Assume AI only automates 50% of legal tasks, and that Jevons' Paradox generates some net new demand. In that case, will there be more lawyers?
Almost certainly not.
Firms do not respond to productivity gains by maintaining headcount and expanding volume. If ten lawyers can now do the work that previously required twenty, it needs demand to grow by 100% just to maintain existing headcount. That is a much harder bar to clear than the Jevons' optimists tend to acknowledge.
And the distribution of what gets automated matters as much as the percentage. The work AI handles best — document review, first drafts, research memos, due diligence checklists, standard contract generation — is exactly the work that currently employs junior lawyers. When that work goes to AI, entry-level hiring collapses. When entry-level hiring collapses, so does the pipeline for training future senior lawyers. The profession loses its replenishment mechanism from the bottom up.
We are already hearing this from firms. The “good enough” threshold has already been crossed for large categories of junior legal work. The headcount consequences are only beginning to show up.
“Good Enough” Is Already Here
The entire defence of the legal profession rests on an assumption: that "good enough" AI is not actually good enough, that clients will continue to pay for human judgment to close the gap, and that there will always be enough premium work to sustain a large profession.
It is worth noting that "good enough" is not a static bar. It is a moving threshold that AI is clearing for more and more tasks, more and more quickly.
A startup founder reviewing an NDA does not need a senior partner. They need something that catches the obvious risks. AI already does that.
A small business signing a supplier agreement does not need bespoke counsel. They need something that won't blow up on them. AI already does that.
A fund processing side letters does not need fresh analysis on each document. They need reliable pattern recognition at volume. AI already does that.
The likely forecast — to be economic brutalists about it — is not that lawyers will migrate upmarket as AI absorbs routine work. It is that more lawyers will lose their jobs every time AI crosses the next "good enough" threshold.
Some legal work will remain — genuinely novel litigation, politically complex regulatory strategy, unprecedented M&A transactions. These require humans who can extrapolate. But this will employ only a small fraction of the current profession, and they cannot absorb the displacement of everyone else at scale. Highly advanced and strategic legal work is not a safe harbour. They are a lifeboat on the Titanic.
