Chief

My teammate Jishnuu Senthil Kumar and I built Chief at the Y Combinator *Call My Agent* hackathon, where we placed 3rd overall out of 100+ teams. Chief is an AI chief of staff for high-stakes conversations, inspired by Penn professor Chris Callison-Burch's research on overhearing agents.

The core idea: in a live conversation, AI isn't useful as a chatbot. It can't sit waiting for a prompt. It has to know when to stay silent, when to surface information, and when to act on your behalf.

Chief listens to your live conversation, builds a running belief of what is happening, and retrieves relevant memory from past conversations. After each utterance, it decides whether to stay silent or act. When the conversation calls for it, Chief acts: surfacing context, drafting the email you just promised, browsing the web for a fact, pulling a mentioned document, scheduling an agreed event, or placing a side call while you keep talking.

That decision, when to help and how, is what separates Chief from AI meeting notetakers, live AI overlays, and generic assistants. Instead of just remembering what happened or showing passive suggestions while you talk, Chief intelligently decides when to step in and can actually take action.

You can watch the demo, read the paper, or browse the code.