Safe
Reliable
Autonomous
We're building AI that can run a megaproject: remembering everything, planning over months, and coordinating across teams.
Horizon: Our Agent Learning Benchmark
Today, we're releasing a preview of Horizon, our benchmark that measures an agent's ability to acquire learnings from a long history and apply them to a task.
Building Long-Horizon Agents
We present a method for building long-horizon agents that work continuously over time, schedule their own activities, and create workflows dynamically. Unlike traditional agents that only respond to user input, long-horizon agents actively pursue goals without constant prompting.
Designing Conversationality in Voice Agents
We explore how to build proactive voice agents that work independently of user input. By flipping the traditional voice pipeline, we create agents that can speak first, handle interruptions, and maintain natural conversation flow.