Trails / Agent Safety and Memory

Agent Safety and Memory

Parts

1. Part 1

   The Memory Problem

   Why AI Agents Forget Everything and What To Do About It

   16 min
2. Part 2

   MCP Is an Interface Boundary

   Why agent tools need typed contracts, policy, and confirmation instead of clever prompt glue

   16 min engineering operations
3. Part 3

   The Injection Problem

   Defending LLM agents against prompt injection when they read the world

   18 min engineering operations
4. Part 4

   The Quiet Teammate

   What Happens When an AI Agent Picks Up the Work Nobody Owns

   16 min
5. Part 5

   Please Do Not Worship the Robot

   A theatrical reminder that Q is useful because the team makes him show his work

   11 min business engineering operations
6. Part 6

   Surviving the Upgrade

   What happens when the ground shifts under a running AI agent — and how three days of misdiagnosis led to a five-minute fix

   18 min engineering operations