Build AI Agents You Can Actually Trust: 4 Foundations

Still explaining every step to your AI—and checking everything it does? In this video, I walk through the four foundations I use to build AI agents I can actually trust: context, skills, visibility, and coordination. This is for anyone ready to move beyond one-off prompts in ChatGPT or Claude and build a system that understands how they work. I'll show examples from my own setup, including agent instructions, reusable skills, tests for those skills, and a place to see what my agents have done. You'll learn: • What belongs in your agent's starting instructions • When to build a skill—and when you're overcomplicating it • How evals help you test and improve your skills • How to get updates without creating another inbox • How to think about the tools you use to communicate with your agents Start small: one clear set of instructions, one useful skill, and somewhere for the work to land. Build trust as you go. What do you wish you could trust an AI agent to handle? Tell me in the comments—and let me know which part you'd like me to explain next.