The Knowledge Gap: Why AI Agents Fail in Real Life

AI agents look great on stage but stumble in real life. They tell you the outdated launch date, suggest writing a follow-up that's already been sent, and keep replying with long texts after you asked for bullet points. What’s missing is proper context engineering, including the background knowledge humans carry without thinking - memories, experience, preferences.

This talk follows Dev Dana, an engineer whose shiny demo agent falls apart the moment it meets real-world workflows. Step by step, you’ll see how Dana:

• adds long-term memory to a chat-only agent, so context from earlier conversations stays available
• uncovers RAG’s role — and its limits
• connects relationship-aware graphs and external data feeds — calendars, docs, email — to keep the agent informed

By the end, Dana’s once-clueless agent navigates projects like a pro. Come for the story, leave with insights you can drop into your own stack to close the knowledge gap for good.