Ambient Agents and the New Agent Inbox ft. Harrison Chase

LangChain CEO Harrison Chase argues ambient agents — event-driven, massively parallel AI — are the next major shift beyond chat agents.

• Ambient agents trigger on event streams, not human prompts, enabling thousands to run in parallel vs. one chat agent at a time.
• Looser latency requirements let ambient agents call far more tools and run explicit planning/reflection steps impossible in real-time chat.
• Ambient does not mean autonomous: Chase identifies four human-in-loop patterns — approve, reject, edit actions, answer mid-run questions, and time-travel to past states.
• Human-in-loop also drives agent memory: without user interactions there is nothing to learn from, stunting long-term improvement.
• LangGraph’s persistence layer stores full agent state at every step, enabling pause/resume/rollback across hours or days.
• LangGraph Platform is being built specifically for bursty, long-running ambient workloads that can spike to thousands of concurrent events.
• Chase runs a personal email agent that has drafted replies and sent calendar invites on his behalf for roughly a year; it is open source on GitHub.

2025-05-07 · Watch on YouTube