Building Defense-in-Depth for AI Agents: A Practical Workshop

What We Will Build By the end of this workshop, you will have a working, layered security architecture for an AI agent. Specifically, we are building a secured customer support bot that uses five independent defense layers to reduce its attack surface from "hope the system prompt holds" to "functionally unexploitable." I am not going to show you how to write a cleverer system prompt. I am going to show you how to engineer a system where a compromised prompt cannot cause meaningful damage. That is a fundamentally different problem, and the solution is architecture, not wordsmithing. Let me show you a pattern I use in every project that ships an AI-powered feature. Prerequisites Python 3.10+ Familiarity with LLM tool/function calling (OpenAI, Anthropic, or similar) Basic understanding of what prompt injection is (attacker tricks the model into following injected instructions instead of yours) A healthy skepticism of "just add a system prompt disclaimer" as a security strategy Step 1: Und