Legal Turing Trap

The loudest promise in legal AI right now is speed. Faster review. Faster drafting. Faster research. This is where the industry keeps pointing its intelligence, and it’s exactly how it walks into what Erik Brynjolfsson calls the Turing Trap.

The trap appears when an industry optimizes for automation instead of augmentation. Automation means replicating what humans already do, just cheaper and faster. Augmentation means creating capabilities that never existed at all. Legal tech companies are overwhelmingly choosing the first path, because it’s easier to imagine, easier to sell, and easier to measure.

This is the legal version of human-like AI: systems trained to behave like junior associates. They read contracts. They draft memos. They summarize cases. Now everyone's a partner, “focusing on high value work”. The theory is simple: swap human for machine, plug into an existing workflow, watch unutilized hours go down. Productivity rises. Value barely moves.

Brynjolfsson’s Daedalus thought experiment captures the problem cleanly. If you automated every job in ancient Greece (shepherding, pottery, weaving) you’d get unprecedented efficiency and still end up with nothing but clay pots and tunics. Nowhere near present day quality of life. Perfect execution of yesterday’s tasks does not invent tomorrow’s industries.

A law firm that perfectly automates today’s workflows will produce digital clay pots and tunics: briefs, contracts, and memos, faster than ever, and fundamentally unchanged. But can we really believe that the law is a game of token crunching?

Firms fall into this trap because automation feels like low-hanging fruit (even though they disregard training costs that are 90% of any AI investment). But still, cost reduction is legible. Headcount math is simple. Replacing labor with software fits existing incentives — tax, accounting, and cultural. Reinventing lawyering, by contrast, requires imagination. It asks a harder question: what legal product and services should exist that currently don’t? What outcomes could the law achieve for people in the age of AI?

That question is mostly being avoided.

Founding Darrow, we asked it first, before we did anything else. We sat on the beach and imagined we did it. What fundementally changed? We could only think of one thing. Unprecedented levels of trust and satisfaction in the legal system. We figured out that humans could never effectively uncover every legal vulnerability and resolve it. So that became our mission.

The real opportunity is augmentation: systems that let humans do things they could never do unaided. Law has its own equivalent frontier, but it isn’t “faster docs.”