In a recent conversation with the managing partner of a major law firm, we got onto a question I hear often: if firms are already using AI responsibly, how much does the business itself really need to change?
His firm is already using it. They have guardrails around it and are thinking carefully about where it fits and where human review is essential.
But he was still somewhat dismissive of the idea that AI would fundamentally change law firms.
One reason he gave was very real.
Judges are increasingly frustrated with legal filings containing fabricated cases, bad citations, and other AI-generated errors. Reuters recently reported that AI-related errors have surfaced in at least 1,395 state and federal court cases. Lawyers have been sanctioned, and the errors have even found their way into judicial opinions and prosecutors’ briefs.
Earlier this year, Sullivan & Cromwell apologized to a federal judge after an AI-assisted filing contained inaccurate citations and misstatements of law. The firm said its own AI policies had not been followed, and a subsequent review failed to catch the errors.
So the managing partner wasn’t wrong about the risk.
Today’s limitations are not tomorrow’s boundaries
What I kept thinking about, though, was how easy it is to look at the weaknesses of an early technology and assume they tell us something lasting about what that technology will be able to do.
Early AI writing had obvious tells. Early systems hallucinated much more. Image generators routinely mangled hands and text. Coding tools were far less reliable than they are today.
Those were real limitations. They just weren’t fixed ones.
The problem comes when today’s limitations quietly become the assumptions on which tomorrow’s strategy is built.
If you begin with what AI can and cannot do today, the questions tend to be operational:
Where can we use it? Where do we need guardrails? What work can we automate? Where do we still need people? How much time can we save?
Those are useful questions. But they are primarily questions about improving the organization you already have.
The more strategic conversation starts somewhere else.
What if many of the limitations we see today continue to diminish? What if the technology becomes significantly more capable, reliable, and less expensive over the next several years?
Then work backward from that future.
What happens to the work itself? Which activities require fewer people? Which forms of judgment become more valuable? How do clients buy the service differently? What happens to pricing, staffing, career paths, incentives, and ultimately the structure of the organization?
A growing business can still face structural disruption
Law firms are particularly interesting because there are very good reasons not to feel a sense of crisis today.
The business is doing extremely well.
Large U.S. law firms reported average revenue growth of 12.4% in the first half of 2026, according to Wells Fargo’s Legal Specialty Group. Demand measured in lawyer hours increased 4.8%, close to the highest level the group has recorded. Billing rates continued rising as well.
And AI may create more legal work, not less.
Companies implementing AI are generating new questions involving intellectual property, employment, data, liability, regulation, governance, contracts, infrastructure, and capital investment. Wells Fargo specifically identified AI-related work as one contributor to the recent surge in law-firm demand.
There is no contradiction here.
AI could create substantially more demand for legal expertise while also changing how that expertise is delivered and who captures the value.
A growing market does not guarantee that value will continue to be created, delivered, and captured in the same way.
That distinction matters well beyond law.
The billable hour is only the beginning
For now, one of the most visible tensions is around the billable hour.
AI can help a firm finish work in a fraction of the time. That is clearly valuable to the client.
But if the firm’s revenue still depends heavily on hours billed, the productivity gain can work against the existing economics.
The American Bar Association has addressed this tension directly. Its guidance makes clear that lawyers cannot charge clients for hours that AI eliminated simply because the work would previously have taken longer.
In most industries, a technology that allows a company to produce the same output with dramatically less labor is an obvious productivity gain.
In a business built substantially around selling professional hours, the relationship is more complicated.
That does not mean the billable hour disappears. It remains useful when the scope and risk of the work are genuinely unpredictable.
But it does mean that “we use AI too” is not necessarily an answer to the deeper strategic question.
If AI materially changes the amount of human labor required to deliver legal outcomes, eventually it puts pressure on the economic system built around that labor.
And billing is only one piece.
What happens when AI removes the work that develops judgment?
There is a second-order consequence I find even more interesting.
Much of the work AI can increasingly compress is junior-associate work: research, first drafts, document review, case chronologies, issue spotting, and due diligence.
Some of that work is tedious. Some of it undoubtedly should become more efficient.
But historically it has done two things at once.
It produced something the client needed, and it helped turn inexperienced lawyers into experienced ones.
The American Bar Association has begun explicitly raising this apprenticeship problem. As AI takes on more entry-level work, firms have to reconsider how young lawyers get the repetitions, exposure, feedback, and proximity to senior judgment through which professional judgment develops.
If a firm needs fewer junior hours today but still needs exceptional senior judgment ten years from now, where does that judgment come from?
The answer cannot be preserving inefficient work simply because previous generations learned that way.
But it also cannot be assuming that someone becomes a trusted adviser because AI got them to the answer faster.
The development model has to change too.
And once you follow that thread, the implications spread quickly.
If less junior labor is required, does the traditional pyramid change?
If clients resist paying for work machines can perform quickly, does pricing change?
If fewer people perform certain tasks, how does leverage change?
If valuable knowledge can increasingly be captured and reused, does expertise remain as dependent on the individual partner?
If clients develop much stronger AI-enabled legal capabilities internally, which work continues to belong outside the organization?
And if a new competitor starts without the existing partnership economics, associate structure, compensation model, and billing assumptions, what might it do differently?
Those are not technology questions.
They are questions about the operating model of the firm.
Using AI is not the same as adapting the business
None of this requires believing traditional law firms are about to disappear.
The strongest firms have tremendous advantages: trusted relationships, institutional reputation, specialized expertise, deep experience, and lawyers whose judgment matters enormously when the stakes are high.
AI could make many of those firms stronger.
It could also make senior judgment more valuable as the cost of producing basic analysis falls.
But that is different from assuming the organization surrounding that judgment remains unchanged.
And this is where the lesson extends far beyond law.
Every organization can buy access to powerful models. Every function can find productivity use cases. Every executive team can put guardrails around usage.
Those things matter.
But the biggest strategic consequences may show up somewhere else.
AI can change the economics of the work. It can change which skills matter and how people develop them. It can change what customers are willing to pay for, which activities belong inside or outside the organization, and eventually the structure of the organization delivering the value.
That is a much bigger leadership challenge than deciding where to deploy AI.
Which is why one of the questions I increasingly encourage leadership teams to wrestle with is not simply:
Where is AI still falling short today?
It is:
If those limitations keep disappearing, what will we need to change about how our organization works?
Then work backward.
Because by the time the answer becomes obvious, many of the hardest things to change quickly may already be deeply embedded in incentives, economics, talent models, structures, and ways of working.
The organizations that gain the most from AI may not be the ones that adopt it fastest.
They may be the ones that recognize earliest how much else it could change, and begin rethinking the business before the market forces them to.