Stop Babysitting Your AI: Context Gathering & Code Review, Looped! (with Alexey Verkhovsky)
Stop babysitting context gathering and code reviews: Quick Dev folds BMAD's create-story, dev-story, and review-story into one agentic loop, and in this episode key open source contributor Alexey Verkhovsky walks the whole diagram with Brian: intent, spec, implement, adversarial review, triage. How it evolved and where it is going. Watch the video version: https://youtu.be/ufOyrZhRj6U Join the BMAD Discord: https://discord.gg/gk8jAdXWmj Everything BMAD in one place: https://bmadcode.com/ GitHub: https://github.com/bmad-code-org/BMAD-METHOD/ Support the show: https://buymeacoffee.com/bmad Chapters: 0:00 Welcome to BMad Code Live 0:40 Meet Alexey Verkhovsky 1:39 Why Quick Dev exists 9:06 Quick Dev vs. vibe coding 14:36 Context windows and human attention 21:18 Brownfield code and the dev-loop question 29:53 How big should an intent be? 36:46 Walking through the Quick Dev diagram 41:26 One-shot mode for small changes 45:13 Plan: the story spec template 51:00 Implement, then never trust the diff 56:54 The adversarial review layers and their cost 1:00:00 The valley of despair: triaging AI findings 1:03:46 You are the bottleneck, not the AI 1:05:59 Customizing the review layers 1:11:39 Q&A: do you even need a PRD? 1:17:30 Q&A: one model codes, another reviews 1:21:28 Quick Dev vs. BMAD Loop vs. Dev Auto 1:23:34 Testing, TDD, and the benchmark hype 1:31:52 Triage, loopbacks, and fixing findings 1:41:05 Deferred work: parking the side quests 1:50:49 Present: a narrative for reviewing the change 1:56:00 Story, epic, or whole app: how much to batch 1:59:48 The story-one tracer bullet strategy 2:04:06 Q&A: running reviewers without sub-agents 2:06:42 Predictions and sign-off