The Jevons Paradox Is Playing Out in Legal AI Right Now
AI was supposed to shrink legal work. Here's why the Jevons Paradox says the opposite is happening at law firms.
A few years ago, the prevailing fear was simple: AI will make legal work faster, clients will pay less, and firms will shrink. It was a zero-sum story: faster output, fewer hours billed, less revenue.
However, that story doesn't appear to be holding up.
What's actually happening looks a lot more like the Jevons paradox, a 19th-century observation about coal: when steam engines became more efficient, coal consumption didn't drop. Instead, efficiency made coal-powered work cheaper, which made people do more of it.
There’s a more modern parallel, in radiology, that’s even more instructive. About a decade ago, a prominent AI researcher predicted that medical schools should stop training radiologists because deep learning would replace them within five years. He was right about the technology; AI did get better and faster at reading scans. But in the years since, the number of radiologists has grown substantially and their salaries have climbed alongside. Why is that? Less expensive and faster scans meant hospitals ordered far more of them. Demand expanded to consume the new capacity and then some.
It’s becoming clear that legal work is heading in the same direction. Demand for legal services has never been static, and we're starting to see the data that bears this out.
Where the reclaimed time actually goes
By most accounts, heavy AI users at law firms are reclaiming significant time each month. The more useful thing to track is what happens next, because not all firms are using that time the same way.
Stage one: speed. The associate produces a first draft in 90 minutes instead of six hours. She moves to the next matter in the queue, the partner reviews it sooner, and then the client gets it sooner. This is table stakes, and it's where most firms are today. The majority of the reported benefit falls here: clients receive work faster, which is valuable but leaves a lot on the table.
Stage two: scope. This is where things get interesting. The associate who drafted the memo in 90 minutes now has time to stress-test the analysis, check an additional jurisdiction, and flag a commercial risk the client didn't ask about but will be glad to know about. The deliverable doesn't just arrive faster. It arrives better. And critically, the expanded scope is still billable. The firm isn't losing hours. It's filling them with higher-value work on the same matter. A 30-point due diligence instead of a 10-point one. A memo that anticipates the follow-up question before the client asks it.
Stage three: net-new work. This is the real prize. The junior associate who flagged the new risks creates a reason for the partner to call the client. That conversation, which previously wouldn't have happened because the partner was spending the afternoon reviewing a draft, generates a new instruction. The associate, no longer buried in routine drafting, picks it up the same week, and the cycle starts to compound.
Most firms aren't consistently at stage three yet. But the ones thinking about it deliberately, asking "where is the saved time going and how do we direct it?", are the ones that will capture the actual economic value of AI. The firms that aren't asking the question are just doing the same work a bit faster. That advantage erodes quickly.
The client side makes this even more interesting
There's a parallel dynamic on the client side. General counsel and in-house teams are deploying AI themselves. Not at the same pace as firms, but it's accelerating, and it changes the demand picture in ways that aren't initially obvious.
Consider what happens when a GC can do a first-pass review in an hour using AI, something they previously would have sent to outside counsel. On the surface, that looks like lost work for the firm. But in practice, the opposite seems to be happening. The GC does the first pass, identifies the three genuinely complex issues, and comes to the firm with a much sharper, more targeted instruction. They're no longer asking for a general review. They're asking for deep analysis on specific points. The work that reaches the firm is harder, more specialized, and more valuable per hour.
This dynamic plays out across the full range of in-house legal operations. A legal team that previously lacked the bandwidth to review a mid-priority contract portfolio now runs a first pass with AI and surfaces the 15% of agreements that need real attention. That 15% goes to outside counsel as a focused, well-scoped engagement. The firm gets a better-defined brief, the client gets a faster turnaround, and neither side wastes time on the 85% that turned out to be clean.
There's also the category of work that simply didn't happen before. When in-house teams are stretched thin, lower-priority matters sit in a queue or get handled with a lighter touch than they probably deserve. Regulatory horizon-scanning, subsidiary governance reviews, vendor contract audits: all the work that's important but not urgent enough to justify outside counsel fees at full scope. When a GC's own capacity expands through AI, that latent demand starts to surface. The work gets scoped, and at least some portion of it flows to firms, often as targeted, high-specificity engagements that didn't previously exist in the firm's pipeline.
The net effect isn't a shrinking pie. It's a different kind of pie, one where the slices are more complex, more targeted, and in many cases more profitable for the firms that are positioned to handle them. The firms that will struggle are the ones still optimized for high-volume, low-complexity review work, because that's exactly the category where in-house AI adoption bites first.
What this means for firms
The firms that will benefit most from legal AI aren't simply the ones that adopt it. They're the ones that understand where the reclaimed time is going, recognize that client demand is shifting toward complexity and specificity, and ask themselves honestly whether they're capturing that value or just running the same playbook a little faster.
The question worth sitting with is straightforward: when your lawyers save time with AI, where does that time actually end up? If the answer is mostly "they move to the next matter in the queue," you're at stage one. That's a fine place to start, but the firms that deliberately push toward stage two and stage three, expanding scope and generating net-new work, are the ones building a durable advantage.
Meanwhile, the firms still optimized primarily for high-volume, low-complexity work should be paying close attention to what's happening on the client side, because that's exactly the category where in-house AI adoption displaces outside counsel first. The work that survives, and grows, is the work that requires real judgment, creativity, and depth.
If this interests you, book a meeting with a member of our team to know more.