Nobody inside your company can un-know how the product works, which is why dogfooding catches breakage and still misses what a real user struggles with. NN/g named that well this week, and it is the easy part. The harder problem sits one step later: when research contradicts the strategy a design was already built around, fixing it properly costs two or three months, so the finding gets logged and the build carries on. The root cause stays in as a leak, and it eventually opens another one. Both frameworks this week are attempts to make that moment survivable — PROVE scores one tool against one recurring task, where three means nothing changed, and Karl Koch refuses to call anything a rule until you can describe the failure it prevents. But the artefact was never the constraint. How fast a team prioritises the follow-up after a feature misses is its real answer.
Nielsen Norman GroupDogfooding vs. QA vs. User ResearchA distinction between three things teams routinely conflate: QA catches reliability and functionality gaps, user research catches usability and comprehension, and dogfooding catches obvious breakage while systematically missing what a real user would struggle with. Employees cannot un-know how the product works.
Pattern
The argument here for a fortnight has been that judgment should be written down and made checkable. This adds the constraint that direction travels in: your own judgment is the least reliable instrument for predicting a stranger's experience of your work.
A five-part test named PROVE, run on one tool against one recurring task rather than on AI in general. Each dimension scores on a five-point scale where three means nothing changed, and output quality is judged against real work the practitioner has already produced.
Pattern
Fifth entry in six weeks on AI spending being asked to justify itself, most recently the tokenmaxxing fad fading as workplaces cut costs. Those entries described the correction. This is the first that hands over the instrument.
A method for converting a design opinion into something an agent can act on. It has four parts: a trigger for when the rule applies, rules for what to prefer and refuse, checks for knowing it is done, and a rejection list of anti-patterns.
Pattern
Third week running on encoding judgment rather than prompting for it, after the directory of published designer skills and UX-Context Design. The vocabulary has hardened from taste and thinking into triggers, checks and rejection lists.
