Plan Before You Prompt: The AI-First Product Loop | Night School with Derek Peters

Most AI builds go sideways for one reason: someone asked the model to invent the plan. Apply to Gauntlet: https://apply.gauntletai.com/register... Access the lesson breakdown here: https://www.gauntletai.com/night-scho... Night School: free live AI engineering classes every Wednesday, 5 PM CT https://gauntletai.com/night-school?u... In this Gauntlet Night School session, Derek Peters walks through the AI-First Product Loop — a five-move workflow that separates vibe coding a slick demo from shipping a product real users actually use. Using a fictional AI task tracker called "Cadence," Derek traces one feature from idea to ship: writing a vision doc, turning it into a PRD, sequencing a roadmap, prompting an agent to build from those docs, and releasing safely behind feature flags. Every prompt and result shown came from his real planning process. Chapters: 00:00 Welcome to Night School 02:13 Intro: building it vs. getting it launched 04:47 Why building products just changed 06:32 The AI-First Product Loop (5 moves) 08:28 Meet "Cadence": the example product 10:43 Vibe coding vs. plan-then-prompt 13:51 Tooling for each stage 15:51 Planning: slow down to go fast 17:55 Writing the vision document 21:37 Turning a vision into a PRD 24:12 Sequencing the roadmap 26:21 The durable context file 30:12 Prompting: you're the tech lead, not a user 35:52 Good prompt vs. bad prompt 37:45 The build loop: ask, review, test, refine 40:33 Four ways it falls apart 42:29 Shipping behind a feature flag 47:08 Generate tests with the feature 49:17 Instrument and measure 50:24 Recap + your three moves 53:44 Q&A — Night School — free live AI engineering classes every Wednesday at 5 PM CT: https://gauntletai.com/night-school Watch past sessions: https://apply.gauntletai.com/ Apply to Gauntlet: https://gauntletai.com/apply