An agent drafts a pricing proposal; the sales rep adjusts the discount from 25% to 30% and adds a note about competitive pressure. That edit is a decision trace. The model’s proposal is a structured prior — what the system thought was right. The human’s modification is the judgment signal — what actually matters that the model missed.
This is the mechanism that closes the gap identified in Decision traces are the missing data layer — a trillion-dollar gap: instead of bolting on decision logging after the fact, agent-mediated workflows make instrumentation a byproduct of how work gets done. As agents insert themselves into more workflows, more judgment is forced to become explicit through edits, approvals, exceptions, and overrides. The strongest objection — that the most valuable judgment lives in intuition and side conversations — is real, but the thesis only requires enough repeated, high-value decisions to become explicit for the system to start learning.
This connects to Revealed preferences trump stated preferences — track what users do, not what they say at the enterprise level: don’t ask people why they made a decision — observe the delta between what the agent proposed and what the human actually did. And it gives Proprietary feedback loops create moats that widen with every interaction a concrete capture mechanism: each agent correction is a proprietary signal that competitors cannot replicate.