Dev Diary 5- Vibe production bug hunts + System One AI models Jev and Laya

Today we look at some production bugs found in Reprise while deployment and sanity check of Reprise by the AI. How we track the issues as tasks and how the agents tackle them. We also look at System One AI models Jev and Laya. How I did a small test with testing them on my Poocha LLM. 00:00 Introduction & Reprise Phase 7 Bug Hunting 01:23 Production Bugs Breakdown: Missing Handles & Gemini Wiring 04:26 Admin Privileges & Timeline Editing Pipeline 05:56 Docker CI 30-Minute Timeout & Subagent Rate Limit 08:37 Bali Trip Plans & Google OIDC Architecture 11:15 System 1 AI Models: Indecision vs Decision Engines 14:15 TypeSafe Jev Playing DOOM in 100ms Software Loops 16:13 Nandakishore's Laya: Autoregressive Decision Models & Primitives 20:02 Fight Club Automaker Analogy & Real-Time Guardrail Moderation 24:24 "Poocha": Building a Kid-Friendly AI on Gemma 4B 26:28 Benchmarking Guardrails: Regex vs Jev vs Laya 31:18 Subagent Review Catches Silent Transcript Bug & Wrap-Up P.S: I forgot to switch screens at 26:28. So you will continue to watch Poocha screen while I yap about regex and subagents.