Debug and profile agentic app experiences with Instruments

A Podkey summary of Debug and profile agentic app experiences with Instruments, from WWDC 2026.Today’s thread is basically this: building with language models looks simple right up until you need it to be reliable. The big themes are testing outputs that won’t come out the same way twice, figuring out what broke when several models are chained together, and using Apple’s Foundation Models Instrument to actually see what’s happening under the hood. There’s also a very real lesson here about tiny configuration mistakes, because one missing tool left an app politely spinning its wheels with no obvious error at all.Why old-school testing stops workingMulti-model apps add hidden failure pointsWhy observability suddenly matters so muchThe tree view that makes the bug visibleHow DynamicInstructions can quietly trap a workflowThe silent failure problemThe performance numbers that actually matterPractical ways to reduce latencyEvaluating prompt quality with real feedback loopsCompatibility and setupThis podcast was created with Podkey. Make your own at https://podkey.fm