A complaint response that used to take an associate 16 hours now takes 3 to 4 minutes. Goldman Sachs deploying Claude for accounting and compliance work got attention for the name on the door, but the work itself is more revealing: parsing rules, applying them to messy facts, exercising the kind of structured discretion that defines professional roles. Harvard Law School's study of AmLaw 100 firms found 100x productivity gains on specific judgment-intensive tasks. And not one firm anticipates reducing attorney headcount.

Those two facts, held together, are the interesting part.

Professional services firms are judgment-rationing systems. The leverage model that defines law, accounting, and consulting exists because senior judgment is scarce and expensive. Partners at the top, associates grinding at the base. Billable hours are the unit of rationing. Partnership structures govern how the surplus from scarcity gets distributed. The whole architecture works because judgment is hard to produce and impossible to copy.

What Goldman's deployment and the AmLaw pilots suggest is that the execution layer of this pyramid is becoming optional. When a complaint response collapses from 16 hours to minutes, the junior hours that funded the economics simply vanish. The judgment embedded in that work got copied, cheaply, by a system that can run the same pattern a thousand times before lunch.

I keep circling back to something, though. "Professional judgment" has always been two things bundled together. Pattern-matching: applying the right rule to the right facts, parsing complexity, synthesizing across sources. And accountability: standing behind the answer with a license and a reputation. Pattern-matching scales beautifully. Accountability cannot scale at all.

Courts are already catching AI-generated bogus citations at accelerating rates, with sanctions exceeding $100,000 in some cases. Pattern-matching running without accountability attached. The response is telling: the UK's Solicitors Regulation Authority approved Garfield.Law as the first firm authorized to deliver legal services entirely through AI, charging per document rather than per hour. The accountability layer still exists. Who provides it and how has been restructured.

The firms sense the unbundling. Non-equity partners now comprise 51% of all partners at Am Law 100 firms, up from 14% two decades ago. The pyramid was already hollowing. Meanwhile, 64% of in-house legal teams expect to depend less on outside counsel as they build internal AI capabilities. Agents accelerate this, making visible something that was perhaps always true: a significant share of what firms sold as judgment was, in retrospect, pattern-matching dressed in professional authority.

The billable hour contained a quiet absurdity. As one practitioner observed:

Efficiency was the enemy of revenue. When the pattern-matching half of judgment scales like compute, abundant and declining in cost per unit, the organizational structures built around its scarcity face something more unsettling than adaptation. They need a new reason to exist.

A harder question sits underneath all of this. How do professionals build the expertise to handle genuinely hard cases if they never work through the routine ones that built their pattern recognition? We're automating the apprenticeship pathway that created the judgment we're now scaling. Nobody has a clean answer for this yet.

Today is Washington's Birthday. Federal courts are closed, government offices dark. The agents, of course, are still running. Professional judgment used to observe business hours because it lived inside professionals. Infrastructure doesn't take holidays. That small fact says something honest about where this is heading, even if where exactly remains, appropriately, a matter of judgment.