Lee Robinson's Stanford CS146S lecture on how always-on proactive agents work

Lee Robinson gave a Stanford CS146S lecture on how always-on, proactive agents are built, using Grok Bot as the example, according to a post by X user meng shao.

Shao says the talk covers the shift from chat assistants to always-on agents, the model changes that made them feasible, the architecture, the harness, context engineering, and future direction, and links to the video with timestamps.

A later post by the same account describes the lecture as a roughly 45-minute breakdown of GrokBot's architecture. It lists sleep-and-wake VMs, a thin client with a single "send to user" tool, Temporal durable workflows, prompt caching, and layered memory and compaction. That post identifies Robinson as a SpaceXAI model training team member and quotes him saying some of these components are being built at SpaceX.