# Uncertainty Expression Is a Shipping Gate for AI Tools ## Claim An AI tool isn't ready to ship until it can tell you what it does when it's uncertain. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds for user-facing AI features where confident-but-wrong output causes real harm or wasted downstream work. - Weaker for low-stakes, easily-verified outputs where a wrong answer costs almost nothing to catch and correct. - Doesn't specify what "handling uncertainty well" looks like (hedging language, confidence scores, refusal) — only that some visible mechanism must exist. ## Evidence Bob Dunn's open-web podcast names uncertainty-handling as the sharpest practical AI readiness test he's found: "if you can't tell me what your tool does with uncertainty, you not ready to ship it" — offered as a decade-plus-earned filter, not a one-off complaint. ## Links - [[3. Resources/Permanent Notes/AI Risk-Aversion Under Uncertainty Rewards Entity Coherence Over Volume]] — because extends; both treat how a system handles uncertainty as a first-class design/evaluation axis rather than an edge case. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in; this is a usable criterion for evaluating any AI feature before adoption. ## Tests / falsifiers - Find a shipped, widely-adopted AI tool that has no visible uncertainty-handling mechanism and no track record of user harm from confident-wrong output — that would weaken the claim as a hard gate. - Compare user trust/adoption between two otherwise-similar tools that differ only in whether they surface uncertainty. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Readwise/Full Document Contents/Articles/Revisiting What Matters Most]] — states the uncertainty-handling shipping test.