Horizontal vs Vertical Agents: Who Wins? | Model Routing: Commodity or Increasing in Value?

Where routing actually belongs, why vertical agents win on evals, and the three-step process LangChain uses to keep coding agent spend from spiraling.

With Harrison Chase

"I really underestimated the demand for inference. Everything just works. This market is so big. There's just so much demand for it." — Harrison Chase

Harrison Chase, on the episode

Harrison Chase runs LangChain, the company most builders reach for when they're trying to get an agent to actually work instead of falling over. The pop framing of agent-building is that it's mostly a model problem: wait for a smarter model and the hard parts disappear. The operating system Chase is actually running is different. Agents fail because they're chains of nondeterministic black boxes stacked on top of each other, and the fix is almost entirely in the harness, the layer of code, memory, and routing logic that sits between the model and the outside world. That reframes where engineering effort, and company value, should go. This protocol pulls the parts of the conversation where Chase gets specific about how to build that harness and how to keep its costs under control.