The Billable Hour Can't Price What AI Actually Changed in Legal Work

The claim, made in a November 9 piece from Above the Law, is narrow but consequential: the only valid way to judge whether AI helps or hurts legal work is to compare the total cost of producing an acceptable final product with AI against the total cost without it. Not the time to draft. Not the hourly rate. The full cost of getting to a finished, defensible work product, including whatever review, correction, and judgment that requires.

That framing matters because most firms and legal departments do not measure that. They measure hours billed, hours saved, or turnaround time on a single task like drafting a contract clause or summarizing a deposition. Those numbers describe one stage of production. They say nothing about what happens when a junior associate's first draft was wrong and nobody caught it for two rounds of review, or when the senior partner's edit time quietly absorbed the savings the drafting tool claimed to produce. The total cost of an acceptable final product and the cost of a fast first draft are not the same measurement.

The piece does not disclose a dataset, a firm, or a dollar figure. It makes an argument about what the comparison should be, not a report of what any comparison found. That is a reasonable thing to say and a different thing from evidence that AI use currently raises or lowers total production cost in actual legal practice. Readers should treat the argument as a standard to apply, not as a result already in hand.

For an operations lead or managing partner deciding whether to expand AI tool use, the practical shift is in what gets tracked before the decision, not after. Current billing systems capture time by task and matter, but rarely capture rework, the number of review passes a document went through, or who caught which error. Without that data, any claim that a tool cut cost or time rests on an incomplete ledger. Before acting on a number showing time saved, a practitioner should ask whether anyone tracked the downstream review burden, and over what population of matters, routine versus complex, the comparison was drawn.

There is also a harder, unresolved piece embedded in the argument: some of what junior lawyers did by hand was not just drafting, it was training. A first-year who spent six hours building a document learned something from the six hours that a six-minute AI draft does not teach. If that training function disappears from the workflow, the long-run cost shows up later, in the judgment a firm no longer has in its fifth-year associates. No firm has published a way to price that.

The number worth demanding, then, is not hours saved this quarter. It is whether the firm can show, across a real sample of matters, that the finished product cost less end to end, counting every hand that touched it and every correction that followed. Until that accounting exists, cost claims about AI in legal work are describing part of the job and calling it the whole one.