Practical notes on building and deploying AI-assisted apps, published only after their claims and sources are reviewed.
WebMCP Is Enabled. Can Chinese Shoppers Find the Same Products?
A proposed paired catalog experiment separates tool availability, retrieval misses, and unsupported recommendations before an agent shopping launch.
A proposed paired catalog experiment separates tool availability, retrieval misses, and unsupported recommendations before an agent shopping launch.
FairStress exposes a gap between an initial answer and resistance to persuasion. A proposed Build Lab experiment tests identity swaps, unsupported pressure, and legitimate corrections separately.
ExtractBench Pro makes whole-document failure visible. A proposed Build Lab experiment tests which review policy catches consequential mistakes within the time a small team actually has.
A proposed Build Lab study separates sign-in, delegated access, action approval, and completion—with matched prototypes, a scoring sheet, and a bounded release decision.
One email a week. Methodology, raw numbers, and the templates we used.
Describe it, ship it to your own domain in one pass. Free to start — no card.