There's a conversation I keep having with law firm leadership, and it always goes roughly the same way. Someone brings up how AI is finally killing the billable hour, and I find myself pushing back — not because I disagree that the hour is dying, but because I think everyone has the causality backwards.
I ran an immigration law firm, Tech Nomads, for years before I got into legaltech, and if there's one thing that job taught me, it's that clients never really cared how long something took. They cared whether the visa got approved. Whether the filing was done right the first time, without a request for evidence coming back six months later to blow up someone's plans. Billing them by the hour for that always felt a little off to me, even back then, like we were measuring the wrong thing simply because nobody had built a better ruler yet. So when people tell me AI is what's finally breaking the billable hour, my honest reaction is: the crack was already there. AI just happens to be standing in the room when it finally gives way.
The pressure nobody blamed on AI
Go back further than any recent AI headline and you'll find general counsel already pushing hard on this. Once legal spend became a line item that got benchmarked the same way marketing spend or IT spend does, “well, that's just how many hours it took” stopped being a satisfying answer to anyone's boss. I remember sitting across the table from in-house counsel, years before any of this AI conversation existed, getting asked for a flat number up front on a matter that genuinely could have gone three different directions. That wasn't an AI-era question. That was just a company that had gotten tired of open-ended invoices.
Legal ops didn't exist as a real job function a decade ago, or if it did, it was one overworked person buried in a general counsel's office. Now it's an entire department at most large companies, and its whole reason for existing is to interrogate outside counsel spend, run competitive RFPs, and push for arrangements that aren't just metered time. That function grew because legal budgets got large enough to deserve their own management discipline — not because a chatbot showed up and demanded it.
And some categories of legal work were commoditizing on their own well before generative AI became a mainstream tool. Standard NDAs. Incorporation paperwork. A lot of routine immigration filings, honestly — the exact kind of work I did constantly at Tech Nomads. Clients had started sensing, correctly, that this stuff was repeatable enough that “we genuinely don't know how long it'll take” wasn't a credible thing to say anymore. Firms were already drifting toward flat fees for these categories years before anyone was talking about large language models in a legal context.
What immigration law quietly taught me about billing
Immigration work is a strange corner of the legal industry to learn this lesson in, because so much of it is procedural rather than adversarial. A visa petition either meets the standard or it doesn't. There's no jury to persuade, no opposing counsel to out-argue — just a form of evidence that either satisfies an officer or gets kicked back. Hourly billing in that context always felt slightly absurd to me, because the client wasn't paying for our time, they were paying for our judgment about which evidence to include and how to frame it, compressed into a document that might take four hours or forty depending on how messy their situation was going in. We eventually moved a good portion of our work to flat fees, not because we were visionaries, but because clients kept asking for it and we got tired of explaining variable estimates to people who just wanted to know what a green card process was going to cost them. What I didn't fully appreciate at the time was how much internal discipline that required — we had to actually know, from our own case history, roughly how long different fact patterns took, instead of just guessing generously and hoping we didn't lose money on the hard ones. That data-gathering habit turned out to matter a lot more than the pricing decision itself.
Where the talent pressure fits in
There's also something that rarely gets talked about publicly: associate burnout, and the fact that hourly billing quietly rewards working slower. Firm leadership has known this for a long time. It's not a secret so much as an uncomfortable thing nobody wanted to say out loud, because the whole revenue model depended on it not being examined too closely. I've talked to enough partners over the years to know this tension long predates any AI conversation — associates leaving because the incentive structure never quite lined up with doing good, fast work, and firms slowly realizing that rewarding hours instead of outcomes was quietly working against their own retention.
What AI actually changes
So if none of this is new, what does AI actually change? I'd argue it changes what's visible, not what's true. Here's the plain version: if a lawyer uses AI to do in twenty minutes what used to take three hours, and still bills three hours, a client is eventually going to notice — especially a client who's using similar tools themselves and has a rough feel for how long the task should really take now. That tension has technically always existed inside hourly billing. AI just makes it large enough, and sudden enough, that firms can't quietly manage it the way they used to.
There's a version of this happening inside firms too, and it's the one I find most interesting because almost nobody wants to talk about it directly. The moment a firm starts using AI for drafting or research, somebody internally has to answer an uncomfortable question: do we bill for the hours we saved, or not? “Yes, business as usual” is an answer a client relationship can maybe survive once, especially if the client doesn't yet have a clear sense of how much faster the work should have gone. It's not a strategy you can run for long, though. Firms leaning on it are trading a little short-term revenue for a client who starts quietly comparing notes with other outside counsel, or shopping around entirely.
The awkward question every firm eventually faces
I think the mistake — and I've made a version of this mistake myself, building legaltech products — is treating all of this as a billing question. It isn't, really. It's an operational question that happens to show up on the invoice. To actually price outcomes instead of hours, a firm needs things hourly billing never forced it to build.
Real historical data on how long work actually takes, broken down by matter type and complexity, is the first thing missing at most firms. Not vibes dressed up as estimates — actual numbers, pulled from real case history, the way we eventually had to build at Tech Nomads once flat fees became the norm rather than the exception.
An actual scoping discipline is the second thing. The ability to say clearly, before work starts, what's included and what isn't, because the forgiveness hourly billing gave you for scope creep disappears the moment the fee is fixed. A lot of firms have never had to develop this muscle, because for decades the answer to “the matter got more complicated than expected” was simply “we'll bill more hours.”
Some tolerance for pricing risk is the third thing, and it's the one I think gets underestimated the most. Once you fix a fee, the firm eats the variance instead of the client. Some matters run under, some run over, and firm leadership has to get comfortable thinking about a matter as something closer to a bet than a metered service. That's a genuine cultural shift for organizations that have spent generations avoiding exactly that kind of risk.
And finally, technology that actually delivers the efficiency being priced in. This is the part where I've watched a lot of legaltech pitches — including some of my own earlier ones, if I'm honest — get ahead of themselves. Committing to a fixed fee on the assumption of AI-driven speed you don't reliably have yet is really just pricing against a capability that doesn't exist. The tools have to actually work, consistently, before the pricing model built around them makes any sense.
Why the pitch gets oversold
It's tempting to sell “AI finally lets you do outcome-based pricing” as if the technology is the whole story. I understand the appeal of that pitch — it's clean, it's exciting, and it gives buyers a simple reason to act now. But it's closer to a precondition than an answer. The operational work — the scoping, the benchmarking, the internal accountability for pricing risk — is still the hard part, and it's the part most firms have simply never had to build, because hourly billing never asked it of them. A firm that adopts AI tools without doing that operational work is going to find itself pricing fixed fees on guesswork, which is arguably worse than the uncertainty of hourly billing, because now the firm is the one absorbing the miscalculation.
Where this goes from here
I don't expect the billable hour to vanish cleanly, and I'd be skeptical of anyone predicting an overnight, industry-wide switch. What I actually expect is a widening split. The routine, repeatable categories of work — the ones that were already drifting toward fixed fees long before this AI conversation started — move there fully, and quickly, with AI as the thing that makes the old math impossible to defend rather than the thing that started the trend. Genuinely novel, high-stakes, unpredictable work stays closer to hourly billing for a while longer, mostly because nobody has good enough data yet to price outcomes on things that don't repeat often enough to build a real track record.
The firms that come out ahead here won't be the ones with the flashiest AI tools bolted onto an unchanged practice. They'll be the ones that were already doing the unglamorous work — the scoping, the historical data, the willingness to eat some pricing risk — long before any of this became a trend piece. That was true of the handful of firms I watched shift to flat fees in immigration work a decade ago, and I don't think it's going to be any different for the firms navigating this shift now. AI didn't start this shift. It's just made it a lot harder to keep pretending it isn't already well underway.