Lessons Learned Building the Ultimate AI Bug Reporter
summary
An AI pipeline called Yak Shaver records screen/audio, converts to MP4, sends audio to Whisper, and uses the ChatGPT API plus Azure context to generate backlog items that attach screenshots, repro steps, acceptance criteria and at-mentions. Recordings take ~30s–1m; manual reports average ~30 minutes, yielding roughly 72 hours saved in 30 days. Production handles ~50–60 edge cases (cancellations, multi-project mentions), battles hallucinations and prompting variability, and fixes reliability with OneDrive audio pipeline, Whisper + ChatGPT API and structured JSON outputs.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:51Speaker background
- 2:34Bug reporting dilemma
- 8:19Live demo: Yak Shaver
- 11:09Transforming recordings to work items
- 13:42Managing bug reports
- 14:48Real-world insights
- 15:34Troy Hunt comparison
- 18:27Feature requests
- 20:55Beyond time savings
- 25:29Lessons learned
- 30:48AI pricing pitfalls
- 32:33Recording methods
- 35:26Edge cases handling
- 37:13Refining AI content
- 41:51JSON output structure
- 45:26AI project architecture
- 46:33AI agent workflow
- 49:05Agent mode demo
- 51:14Understanding MCP and its role
- 53:45Live demo: Jira + MCP
- 56:55MCP security
- 57:39Future roadmap
- 59:29Final results
- 1:00:10Q&A
related talks
35mThe Podcast Pipeline Nobody Sees: Behind the Scenes with AI Agents and AI gen
01h 04mShaving the Yak with AI - Devs want specs, Product Owners want speed
33mDesigning Voice Agents for Real Conversations
22mAI in User Research: Tools, Use Cases and Best Practices
22mBeyond Bugs: How AI turns QA into Full-Circle Engineers
58mFrom AI Sceptic to AI Sceptic Who Ships
13mMeet the Speakers
49mHow to be an AI First company