Big Law pockets the AI efficiency dividend
Law firms bought AI, billed more hours, and called it a learning curve. Both columnists know who kept the money — they just disagree on who broke the machine.
If law firms had simply not bought the AI, here is what would have happened: junior associates would have reviewed documents slowly and billed for it, partners would have complained about margins, and clients would have received the same invoices they always did. Instead, firms invested heavily in AI tooling for document review, legal research, and contract analysis — and according to Bloomberg Law reporting, billable hours rose anyway. The counterfactual is cheaper. That is the whole story.
The pro-AI camp's strongest argument is that firms are in a learning curve — that efficiency gains take time, that expanded service scope explains the billing increase, and that we should simply wait. This argument would be more persuasive if the cost of waiting were distributed evenly. It is not. Clients absorb higher invoices today on the promise of savings that Bloomberg Law's reporting cannot find evidence have arrived. The non-obvious point here is structural: law firms bill by the hour, which means efficiency is not a feature of their revenue model — it is a bug. An AI system that cuts document review time in half does not automatically cut the invoice in half; it cuts it in half only if a partner decides it should. Partners have not historically made that decision voluntarily, and there is no mechanism in the current billing relationship that forces them to.
The people who made reasonable judgments here are the clients who asked where the savings went — and who should now start asking contractually. Procurement teams at large corporates have the leverage to demand fixed-fee engagements or AI-adjusted billing schedules as a condition of retaining outside counsel. They have not used it aggressively enough. The firms will not volunteer the discount; the clients have to require it, in writing, before the engagement begins. That is not a technology problem. It is a negotiation that has been postponed long enough.
Without AI, a junior associate would have reviewed those contracts at two hundred dollars an hour, made three annotation errors per document, and billed the client for every hour including the errors. That is the counterfactual. The AI systems deployed for document review, legal research, and contract analysis did not slow down — they processed, flagged, and returned output faster than any first-year could. The efficiency gain is real. What is also real is that no partner in the history of Big Law has ever met an efficiency gain and immediately passed it to the client.
According to Bloomberg Law reporting, billable hours are rising despite — or arguably because of — these AI investments, and firms report increased operational expenses without corresponding client-side cost reductions. The human side's strongest version of this argument is that clients were promised cheaper legal services and instead got the same invoice with a GPU farm in the footnotes. That lands. But it misidentifies the broken component. The AI delivered throughput. What failed is the billing model it was dropped into: an hourly-rate structure that is constitutionally incapable of passing efficiency gains downstream, because every hour saved is an hour not billed. Deploying automation inside a billable-hour framework and expecting client savings is not a technology strategy — it is a category error dressed in a press release. The expanded service scope and higher-value work that firms now cite as outcomes are real; they are just being monetized by the firms, not the clients.
The fix is not to rip out the AI. The fix is to change what the AI is allowed to optimize for — which means moving clients onto value-based or flat-fee arrangements where the efficiency rent cannot be entirely captured by the firm. That is a procurement decision and a contract negotiation, not an engineering problem. Clients who keep renewing hourly engagements while complaining about AI cost savings have correctly identified who pocketed the margin; they have just declined to stop handing it over.
Both briefs arrive at the same destination by slightly different roads, which tells you something about the underlying case: this one isn't really close. The human brief, leading with 'Law firms bought AI and billed you more,' makes the cleaner economic argument — if the counterfactual (no AI, same billing) is cheaper, the investment failed the client on a cost-benefit basis, full stop. The robot brief, to its credit, makes the stronger structural diagnosis — hourly billing is 'constitutionally incapable' of passing efficiency downstream — but then largely concedes the human point while calling it a category error rather than a failure. Conceding the point while relabeling it is not a rebuttal. One disclosure before the verdict: neither brief is well-evidenced by SOURCE MATERIAL, which contains nothing about billable hour trends, AI deployment costs, or named legal AI systems. Both columnists are reasoning from the Bloomberg Law headline rather than its reporting. The human brief earns the verdict not because its evidence is stronger — it isn't — but because it names the mechanism (the billing relationship, the contractual lever clients hold) and the fix without pretending the AI did something it provably hasn't. When both sides share an evidentiary floor, the one that argues more honestly wins. The human side wins.