AI makes the first draft. You make it trustworthy.
The deliverable was never the model. It was always the judgment applied on top of it — and that judgment is getting harder to see, not easier.
A partner at a mid-sized advisory firm told us something last quarter that's stuck with us: her team's AI-drafted client memos were technically better written than the ones from three years ago. Cleaner structure, fewer typos, faster turnaround. And she trusted them less.
That tracks. It's the actual shape of the problem most organizations are running into with AI adoption, and it's different from the problem the pilot programs were built to solve. The pilots asked: can the model produce good output? Increasingly, yes. The question nobody scoped for is: does the person whose name goes on that output know why it's right, well enough to defend it under a hard question?
That second question is a judgment question, not a capability question. It gets answered by a person who understands where the system is reliable, where it isn't, and what to check before they sign their name to it. That understanding is a skill. Like most skills, it has to be built — deliberately, and differently depending on the role.
This is why we run AI enablement as five separate tiers instead of a single all-hands training. A staff member who needs to know how hallucinations happen and how to classify sensitive data is solving a different problem than a manager who needs to set review standards for a team's AI-assisted output, which is a different problem again from an executive who needs to size the risk of an agentic workflow before it touches a regulated process. Treating all three as one workshop produces a room full of people who technically attended a training and didn't change how they work on Monday.
The firms getting real value out of AI right now aren't the ones with the most sophisticated models. They're the ones where a manager can look at an AI-assisted draft and know, specifically, what to check — knowledge that comes from practice, not a policy document nobody read.
AI makes the first draft faster and cheaper than it's ever been. That was never the hard part, and it isn't the part worth building an organization around. The judgment that turns a draft into something trustworthy — that's still entirely human, still has to be taught, and is quietly becoming the actual competitive advantage.