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.

Ambient agents have less strict latency requirements compared to chat agents, as they can run for longer periods before needing a response.

Unlike traditional chat agents that require human prompts, ambient agents are triggered by background events, allowing for more simultaneous operations.

Ambient agents listen to an event stream and act on it accordingly, potentially responding to multiple events simultaneously.

We envision a scenario where hundreds of thousands to millions of AI agents are working in parallel, but the real limitations will be in taste and computational resources, not the number of agents.

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.

With the background agent, we aim to simplify the process for developers to create environments where the model can run tests and make it repeatable, allowing for quick updates as code states change.

An example of an ambient agent is an email agent that drafts responses or sends calendar invites, maintaining human oversight through the agent inbox.

The creation of agents is becoming more accessible, allowing individuals to build numerous agents to streamline workflows, similar to creating spreadsheets.