For most of the modern era, an expert was someone who held knowledge other people could not reach. The doctor knew the diagnosis. The analyst knew the comps. The lawyer knew the precedent. Expertise was, in large part, a function of access and recall — you paid for the contents of a head you could not replicate.
That definition is quietly collapsing. A language model can now recall, summarize, and reason across more written material than any human will read in a lifetime. The contents of the head are no longer scarce. If your value was knowing the thing, the floor just moved.
Knowing was never the hard part
The uncomfortable truth is that recall was always the easy half of expertise — it was simply the half we could test for, credential, and bill. The harder half never fit neatly on a résumé: judgment under uncertainty, knowing which question is the real one, sensing when the textbook answer is about to be wrong.
AI is extraordinary at the first half and structurally weak at the second. It can tell you what has been written. It cannot tell you what is true in a specific room, on a specific deal, for a specific patient — the parts that were never written down because they live in someone’s experience.
The moment knowing becomes free, the value migrates to verifying, judging, and deciding — the things knowing was always standing in front of.
The new line: what can be verified vs. what must be asked
We think the real boundary AI is drawing is not human vs. machine. It is verifiable vs. tacit. On one side sits everything a model can ground in a primary source — filings, transcripts, papers, prior records. On the other side sits everything that only exists inside a practitioner: what actually got negotiated, why the trial really stalled, how the system behaves when it is under load and no one is watching.
A machine that is honest about that line is far more useful than one that pretends the line is not there. It should answer confidently from sources, label what it can only infer, and — critically — stop and route to a human for the part that no document contains.
Expertise is being concentrated, not erased
The fear is that AI dilutes expertise. We think it does the opposite: it concentrates it. Once the machine handles recall and first-draft reasoning, what is left for the human is the densest, least substitutable part — the 15 minutes of an expert’s knowledge that no corpus can contain.
That reframes the whole economics. You no longer pay an expert for a day of their general knowledge. You pay for the specific gap only they can fill, exactly when the work hits it. The expert becomes more valuable per minute, not less — because every minute is spent on the part that matters.
The definition of expertise is not shrinking. It is being sharpened to its essential edge: the judgment that survives after everything knowable has already been known.