Vibe Coding vs. Enterprise Software: Where the Security Line Really Is
Most people using AI to build software are not thinking about security, and neither is the AI.
Topics Covered
Show Notes
Most people using AI to build software are not thinking about security, and neither is the AI. Nir Valtman, co-founder of Arnica and former CISO, breaks down why that is a serious problem and what you can do about it today. Arnica helps engineering teams secure the entire development lifecycle by injecting security requirements directly into the tools developers already use. In this conversation, you will learn: - Why AI coding agents repeat vulnerable patterns from your existing codebase, and how to stop it - How to use Claude.md, Copilot instructions, and Cursor rules to enforce secure-by-default code generation - What prompt injection is, how attackers use it against AI agents, and why most users have no idea it is happening - The real security tradeoffs of using autonomous tools like OpenClaw versus sandboxed tools like Cowork - How Anthropic’s Opus 4.6 model is changing the bar for reasoning and security detection - Why vibe coding an MVP is very different from vibe coding enterprise software, and how to tell the difference - A practical prompting tip that uses AI to interview you before building, so nothing important gets left out If you are building with AI tools or running a team that is, subscribe so you do not miss more conversations like this one. Connect with Nir Valtman: Website: arnica.io LinkedIn: search Nir Valtman
Topics Covered
- The Web Talk Show