Inside Anthropic’s Bet on Claude Agents that Work While You Sleep | Jess Yan
A practical look at how Anthropic PMs build and use Claude agents to understand code, monitor feedback and analytics, prep for meetings, and get work done overnight.
TL;DR
- Jess Yan, product lead at Anthropic, shared insights on building long-running AI agents.
- Anthropic teams are transitioning from prompting AI to building agents that can work overnight.
- A demo showcases building a Claude analytics agent from scratch.
- Agents are used internally by Anthropic PMs for understanding the codebase and monitoring customer feedback.
- Features like Traces help identify where agents get stuck, and Evals allow agents to grade their own work.
- Anthropic PMs utilize agents for at least five product management workflows.
- One application involved processing a 4,000-company waitlist using agents.
- A tool called Evan is introduced for preparing users for high-stakes meetings.