SC
Ambient Agents and the New Agent Inbox ft. Harrison Chase
@ Sequoia Capital
05/07/25
Takeaways
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Despite being ambient and operating continuously in the background, responding to events rather than direct human prompts, these agents are not fully autonomous; human interaction is still necessary for approving or rejecting certain actions they take, which sets them apart from traditional chatbots.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Ambient agents listen to an event stream and act on it accordingly, potentially responding to multiple events simultaneously.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Unlike traditional chat agents that require human prompts, ambient agents are triggered by background events, allowing for more simultaneous operations.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Ambient agents have less strict latency requirements compared to chat agents, as they can run for longer periods before needing a response.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
The user experience for ambient agents differs from chat agents, as they operate in the background without direct user interaction, yet it's important for users to engage with them.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
An example of an ambient agent is an email agent that listens for incoming emails and responds or schedules meetings accordingly.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Ambient agents allow for scalability, enabling thousands of them to run in the background, which significantly increases our impact.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Ambient agents can handle more complex operations than chat agents due to their relaxed latency requirements, allowing them to call multiple tools and perform intricate tasks.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Having a human in the loop improves results by allowing agents to ask clarifying questions, leading to better outcomes than relying solely on initial prompts.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Incorporating human oversight builds trust, especially when agents perform explicit actions like sending payments or approvals.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Human interaction with agents enhances their memory by providing user interactions that inform future performance, making them more effective over time.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
The 'agent inbox' prototype allows users to see when agents require actions, providing detailed descriptions and facilitating user interaction with the agent's tasks.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Langraph, the agent orchestration framework, is designed to support ambient agents by ensuring persistence of state, allowing users to modify agent actions at any time.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
Building Langmith focuses on providing visibility and observability for long-running agents, which is crucial for correcting mistakes and managing complex tasks.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
An example of an ambient agent is an email agent that drafts responses or sends calendar invites, maintaining human oversight through the agent inbox.
HC
Harrison Chase
05/07/25
@ Sequoia Capital
The open-source email agent is available on GitHub, showcasing how various components of ambient agents can work together effectively.