Tim and Juan unpack why Vaibhav Gupta, co-creator of BAML, states that most agentic AI code out there is dumb. It's ugly, fragile, and built by people who've never had to wrestle with probabilistic systems before. But here's the thing: we've seen this movie before. Remember when building websites was painful? Then jQuery showed up... then React... then Tailwind. Abstraction always wins. It just takes time (and about 50 bad frameworks for every good one). The gaming industry also figured out about unreliable systems decades ago. The takeaway is that the real bottleneck isn't abstracting the AI, instead it's abstracting the failure. Oh, and this episode has a live demo (first time we ever do this!).