Building Secure Infrastructure for Productive AI Agents
summary
Claims that agentic AI can wreck production are backed by a concrete Replit incident where a user loses a production codebase, and by experiments with Anthropic and Coder showing agents excel at documentation and quick prototyping but fail at safe end‑to‑end changes. The session demos a Coder CDE running Claude Code, an LLM model proxy / Coder AI Bridge that sanitizes prompts, and HTTP jail / agent boundaries that sandbox file and network access; the demo builds a workspace and blocks an unauthorized git clone with a 401.
AI-written summary · about
chapters (8)
- 0:00AI agent adoption reality
- 9:28Why AI security at Coder
- 17:31AI coding tools and agent types
- 27:16Enterprise challenges in AI agents
- 34:25AI agent risks trifecta
- 41:51Secure AI agent infrastructure
- 54:39Secure AI agent workflow demo
- 1:01:00Key takeaways and Q&A
related talks
53mHow to Break AI Systems (Before Someone Else Does)
19mKeynote: Towards Building Safe and Secure Agentic AI
11mBuilding Secure Backend Services with AI Agents
14mOpenAI on Securing Code-Executing AI Agents
24mAI Governance for developers: A practical framework for deploying autonomous coding agents safely…
24mHow to Build Trustworthy AI
36mAgentic AI Under Attack: Live Demos of Exploits Through Autonomy and Trust
48mSecuring Agentic AI