The Gold Master Problem
A consultant wrote something true this week, and he happens to sell the fix.
Rob Hanna runs a content consultancy out of Toronto. He wrote a piece for Newsweek arguing that companies are getting AI wrong at the foundation. Not the model. Not the compute. The knowledge underneath it.
His point, stripped down: businesses have spent decades treating financial data and customer data like something that needs rules, ownership, and review. Nobody did that for the actual written knowledge sitting inside a company. The manuals. The policies. The internal playbooks written by three different people over ten years, half of them outdated, none of them marked which one wins when they disagree.
Feed that mess into an AI system and the AI doesn’t know it’s a mess. It just answers with whatever it found, confident either way. Hanna’s line on this is the one worth keeping: bigger models don’t fix that problem. They just produce faster, more confident versions of the same confusion that was already there.
His fix is what he calls a gold master. One trusted source, owned by somebody, reviewed on a schedule, updated when things change. Everything else gets built around that instead of just dumped in alongside it.
Worth saying plainly: Hanna sells this. His company does exactly the work he’s describing the need for. That doesn’t make him wrong. It means the argument needs to stand on its own, not on his say-so, because he’s got a reason to want you to believe it.
It stands on its own.
This is the same failure we already named in a different room. A week back, this page laid out the empty seat inside multi-agent AI systems — the orchestrator that routes tasks between agents but never checks anything on the way through. That piece was about who’s accountable for an action once a system already decided to take it.
This is the step before that. Before an AI ever acts, it has to know something. And if what it knows is scattered across a dozen documents that quietly contradict each other, with nobody assigned to say which one is true, then the AI was never going to make a sound decision in the first place. Not because it’s a bad model. Because nobody owned the truth it was standing on.
That’s the same principle running through everything this page has stood on since day one. Responsibility for a decision with real consequence has to stay explicitly human, traceable, and owned at the moment of commitment. Hanna’s arguing for that at the knowledge layer. We’ve been arguing for it at the decision layer. Different floor of the same building.
Here’s the balance worth naming instead of skipping past: neither one of us has the whole fix by itself. A gold master document doesn’t stop an agent from acting outside its scope once it’s moving. A governed orchestrator doesn’t help if what it’s acting on was garbage to begin with. You need both ends held, the knowledge and the action, or the well-governed half just hides the broken half a little longer.
Nobody sent this argument looking for a customer. It showed up because a Toronto content consultant ran into the same wall from the business side that this page has been naming from the governance side. That’s what convergence actually looks like — two people who don’t know each other, coming at the same problem from opposite doors, and landing on the same word: ownership.
© 2026 The Faust Baseline LLC | All Rights Reserved
