20VC: Harrison Chase on Inference Demand, Routing, and Why Agents Break

Inference isn’t getting cheaper or simpler — it’s becoming the hidden battleground for AI products. In this condensed summary of 20VC, host Harry Stebbings speaks with LangChain CEO and founder Harrison Chase about the exploding demand for inference, why agents are so hard to build, and where the real value in model routing and vertical agents may lie. This shorter recap turns the full conversation into a quick listen, focusing on the key takeaways: why non-determinism makes agents fragile, how observability and evals help catch failures, why companies should think about portability and "owning your intelligence," and why routing is not just a commodity. Chase also explains how domain expertise, harness design, and internal traces matter more than generic benchmarks, especially as coding agents and inference costs keep rising. Listen now to get the key ideas in minutes.