Experts are AI's
Complements.
Not its Bouncers.
AI is unreliable, so you need experts to babysit it. That makes domain knowledge a safety net on top of flaky AI — an argument that weakens every time the model gets better.
"I think the opposite is true: domain knowledge gets more valuable because horizontal LLMs are good — not in spite of it."
"How would you contrast your reasons why?"
value moves to its complements." — Jordan Kyriakidis, CEO, QRA Corp
The Three Arguments
Cheap compute made "people who know what to compute" valuable. Cheap general intelligence makes "people who know what intelligence to point at what problem" valuable. Experts are not getting replaced — their leverage is going up, because the thing they direct is now wildly more productive per unit of their attention.
The LLM can "build" almost anything. Knowing what to build — which tradeoffs, for whom, under which constraints — is not in the training data. "Build me a CRM" will get you generic slop. The interesting prompt is one only a domain expert could even write.
LLMs have absorbed the codified layer of every field. What is left is the stuff never written down: intuitions, "we tried that in 2019 and here is why it broke," the unspoken constraints. As models eat the codified layer, the relative value of the tacit layer grows.
They are AI's complements. More valuable
as the model gets more capable — not less." — Jordan Kyriakidis, CEO, QRA Corp
At QRA Corp, we have deep domain knowledge in multiple areas. The challenge is how to codify that domain knowledge in a way that makes a progressively more powerful LLM more valuable — not less.
In the Scribe and Gavel framework, the Gavel is where the domain knowledge lives. The Gavel compounds as horizontal LLMs get better. It is not a safety net. It is the source of leverage.
Expert leverage increases as AI capability increases — not the reverse.