A non-technical product manager named Bryce Rattner Keithley built and shipped a working iPhone app to the App Store using only AI tools, no coding background, and no engineering help. This is the central case study in a Lenny's Newsletter podcast episode that doubles as a practical field report on what vibe coding actually looks like end to end, including the failures, the workarounds, and the specific tools used.
The episode also covers two other distinct threads worth tracking: an explanation of Codex Goals, OpenAI's framework for steering agent behavior in its CLI coding tool, and a hands-on review of Claude Opus 4.5, Anthropic's latest frontier model. These are not surface-level takes. The Codex section gets into how goal specification affects output quality in agentic workflows, a problem that matters more as autonomous coding sessions get longer and less supervised.
Read the full piece if you want the specific toolchain Keithley used, where the process broke down, and how he debugged issues without reading a single line of code. The broader implication is direct: the bottleneck for shipping software is shifting from technical skill to taste and problem definition. That shift is already underway.
[READ ORIGINAL →]