AI won't kill the wizards

“AI will kill all the lawyers” is the kind of headline that travels fast because it flatters a simple story. The Spectator piece does exactly that. It gestures at real capabilities, then leaps to an absolute. The problem isn’t that AI is useless for lawyering. It’s that usefulness is being mistaken for replacement.

The work begins where rules stop being obvious. I believe issue spotting is the core activity, and it is not only pattern recognition. It is creative. It requires imagining fact patterns that have not yet appeared, sensing tension between principles before they openly collide, and noticing what the text isn't saying. The world isn't a chess board. It does not pause long enough to be fully learned so that the next issue be spotted without a human heart.

Specialized models can do some of this. They already do. But they do it in bounded spaces, with curated assumptions, and with a kind of statistical courage that breaks down at the edges. Catching all the issues—across domains, technologies, evidentiary regimes, and social change—is a fundamentally open-ended problem. Training, continuously updating, and maintaining a machine that can reliably spot issues, evaluate their merits and impact, and resolve them correctly requires humans.

The irony is that AI does not simplify law. It makes it even more arcane. The likely side effect of AI-augmented lawyering is a deeper wizard-priesthood in lawyers and lawmaking and enforcing. Legal reasoning now fuses with domain expertise — e.g., privacy means data architecture — layered with software and AI literacy, and capped by evidentiary mastery in both collection and analysis across the various types of sources. Law drifts away from a rulebook and toward an even more complicated operating system. Powerful, flexible, and unreadable to most people who are subject to it.

That is the real challenge hiding behind the hype. Not whether AI kills lawyers, but whether it locks the system further away from normal citizens. The question worth asking is how to give people the ability to understand the legal machinery that governs them, to navigate it autonomously, and to act without needing a wizard to translate the runes. That problem is harder than automation. It is also far more important.