Harrison Chase says proactive AI agents raise unsolved context engineering problems

LangChain's Harrison Chase says agent builders are converging on a standard architecture: an LLM running in a loop and calling tools inside a sandbox.

He argues that proactive, always-on, goal-oriented agents create new context engineering problems, including when to notify users, what memory to keep, and how to handle asynchronous subagent work. Chase says LangChain and the industry do not yet have general solutions to these problems.