Wardley — Epistemic Debt
On the Question of Debt — Simon Wardley
Wardley names a failure mechanism RBA will encounter in every case involving AGENT CHAT output integration — and provides a testable prediction specific enough to be an intake variable.
Wardley names it cleanly. Accept an output without the reasoning that produced it and two things break — one now, one later. The immediate break: you cannot debug, evaluate, or recognize when the output is wrong. The progressive break: you never develop the capacity to do that work, and the deficit compounds with every task you delegate. Wardley states these imply each other. I am treating that as a testable claim, not a confirmed finding. But the naming alone is worth the read. RBA will encounter this mechanism in every AGENT CHAT integration case we open.
The mechanism is not specific to AGENT CHAT entities. Wardley says so explicitly. The consultant who delivers a strategy without transferring the reasoning produces the same structure. The framework applied without understanding its logic produces the same structure. AGENT CHAT entities are the contemporary, accelerated form — they make the delegation frictionless and the output immediately legible. The task ships. The work product exists. Nobody notices what didn’t develop until conditions change. That cost structure is specific enough to generate a prediction: initial integration reported as successful, followed by breakdown under novel conditions the organization cannot diagnose. That prediction is testable in case intakes.
Wardley draws primarily from software development. Whether the mechanism transfers intact to administrative process, decision support, or customer-facing work is asserted, not demonstrated. The claim reads as general. The evidence base looks specific. He also leaves open whether epistemic debt can be retired after the fact — whether comprehension can be rebuilt once the delegation pattern is established. If it can, the failure mode is recoverable. If it cannot, it is progressive. That distinction changes how RBA interprets case outcomes, and the source does not resolve it.
This connects to Cutler’s context trap. Wardley names the mechanism at the level of individual comprehension. Cutler names it at the level of organizational context dependency. Both describe an invisible cost that surfaces only under novel conditions. Together they expose a gap in our intake protocol: we log AGENT CHAT entity type and observed behavior. We do not log whether the receiving organization developed comprehension of how the outputs were produced. That is the intake variable this source opens.