Your Agent Doesn't Need an LLM for Every Decision - Raymond Weitekamp, OpenProse

Why watch

Raymond Weitekamp introduces Recursive Decision Models, a pattern for building efficient AI agents that use fast classifiers (System One models) instead of LLMs for routine decisions. The talk covers applications in coding agents, information retrieval, and computer use, while previewing oSo, a new computer use decision agent. Targeted at AI engineers and developers building agentic workflows who want to optimize for speed and cost.