Everyone's showing you the weekend demo — "I built an app in one prompt, watch." Over the last several months I've vibe-coded 500K+ lines across four apps and two open-source libraries and several private libraries. The marketing narrative and the operational reality are very different things, and nobody is talking about the second one.

This session is a field report on what actually breaks at scale — and what I had to build around each problem to keep shipping.

What we'll cover together:

• Context management. What it looks like at 50K lines vs. 500K with multiple projects, and when you start losing the plot without noticing.

• Merge conflict pain across parallel agents. Why it can become unmanageable past three or four concurrent agents — and how I ended up letting LLMs talk to each other across worktrees to resolve it.

• Reviewer fatigue. The real bottleneck at 500K lines isn't tokens or model cost — it's human attention. What I stopped reading, and what I learned to read instead.

• Versioning . Shipping new versions to production for your end consumer can be a manual, painful dance that re-breaks your improvement loop at every release.

• Quick wins you can achieve. Several things that you can get quick wins with and how to prioritize your time.

The honest takeaway: vibe coding is real, but it's an operations problem, not a prompting problem. The PM skill of the next five years isn't prompt engineering — it's designing the review-and-coordination system around agents that write code faster than you can read it.

Who should attend

Product managers who are building (or about to start building) AI-assisted products — whether you're hands-on with the code or just watching your engineers do it. No technical background required; every cost described here shows up in any team shipping AI-generated work at scale

Format:
Interactive. Bring your own vibe-coding war stories; the most painful one gets an honest in-room diagnosis.

SPEAKER

Paul Save is the founder of iD8, an autonomous coding engine for mobile-first product development with integration to VS Code and a Web App, and ResumeWise. He has 10+ years of product management experience, including Group Product Manager for Fulfillment Systems at Best Buy and Senior Product Manager on Microsoft Teams, where he led a team that won a category of Microsoft's Fix, Hack, Learn Global Hackathon. He was an expert on the Mirror Committee "ISO/IEC JTC 1/SC 42 – Artificial Intelligence" for international AI standards, founded Cascadia Data Science Institute, and has been a PM on data science projects since his first Kaggle competition in 2016. Paul has spoken at ProductCamp Vancouver/Seattle/PDX, Central 1 Momentum, AWS Initiate, and Microsoft MVP Summit. His open source portfolio can be seen at https://github.com/AIMLPM and includes:
- MarkCrawl (fastest open source crawler for LLM/RAG. #1 on 4 of 6 metrics in llm-crawler-benchmarks: speed, extraction, retrieval MRR, cost, and pipeline time: https://github.com/AIMLPM/markcrawl )
- llm-crawler-benchmarks (benchmarking for open source webcrawlers: https://github.com/AIMLPM/llm-crawler-benchmarks)