Claude Cowork makes MCP matter again. For about two years, AI folks have been pushing the importance of connectors and MCP – model context protocol – as the way that agents can communicate with legacy services via API. To be clear, APIs still matter a great deal. But for the non-technical user, APIs are a…
Summary In today's episode, I explain how to protect AI agents from trusting misinformation and disinformation. Here's what this means for you. You can build more reliable and secure agents by implementing guard models and specific source parameters. You'll also learn these concepts: how guard models identify bad patterns, why AI treats probability as truth,…
In the world of product market fit, there are three levels: Done for you: least work, most cost Done with you: 50/50 split on work, medium cost Done by you: most work, least cost The analogy I often use: Done for you: dining out. None of the work, all of the cost Done with you:…
Summary In today's episode, I explain how to navigate bot detection and website blockers when building production AI agents. Here's what this means for you. You can build more resilient agents by treating web access as an infrastructure challenge rather than an AI intelligence issue. You'll also learn these concepts: how specialized libraries like Playwright…
Summary In today's episode, I explore ways to ensure reliability and debugging when building AI agents and workflows. Here's what this means for you. You gain the tools to create more stable and predictable AI systems. You'll also learn these concepts: how logging reasoning processes improves auditability, how to implement global and task-based guards, and…
Of all the models DeepSeek has released lately, none is more interesting to me than DeepSeek Math V2. Not because I have a particular affinity for math. I really don’t. It was the subject I did worst in school. (I credit my 5th grade teacher for instilling a deep seated lack of confidence in me…
The AI benchmark to keep your eye on in 2026? RLI. RLI, or the Remote Labor Index, published by the Center for AI Safety, is a fascinating, real world benchmark. AI is given a project that a remote worker would normally do, valued in the thousands of dollars and hundreds of hours of work for…
Summary In today's episode, I debunk the idea that AI models possess consciousness or sentience. Here's what this means for you. You gain clarity on why the debate over machine sentience distracts from real-world AI risks. You'll also learn these concepts: how AI functions as a mere database of probabilities, how digital files differ from…
Here’s 2 people fully harvesting 30 acres of wheat in an hour. That’s a John Deere X9 1100 Combine Harvester. It’s gigantic, the width of a small house. It cuts, threshes, and winnows all in one step, pouring the finished grain into a grain hopper. For those unfamiliar, harvesting wheat is a pain in the…
