a16z is backing First Person Labs to build Second, a continuously updated model of you for the agents you choose.
Runs on your device by default.
That default is the tell, in the pitch from a16z partners Joel de la Garza and Martin Casado. Second uses a background engine to build context over time, then serves it across apps instead of trapping it in each one. The model is inspectable and portable. You choose the intelligence that powers it and what other applications can access, according to the firm.
You have felt the break. You teach one assistant your calendar rules, then you start cold in the next app. Second wants that file to travel with you, not rebuild from chat logs each time.
The bar is already moving. Per The Information, ChatGPT stores user details and lookups while Claude Code keeps auto memory notes on preferences and feedback. First Person Labs is arguing those memories do not count if you cannot see, move, or cut them off.
The team fits the job. Ken Liu researched AI and data privacy at Stanford and at Google DeepMind, including what models remember and how to protect it, per a16z. Erik Chi worked on network security, privacy and censorship resistance at Michigan. Together they built the Open Anonymity Project for more private AI access.
On-device still has sharp edges. Permissions have to stay current, sync has to work, and inspectable only helps if you check before an agent shares too much. If First Person Labs gets that workflow right, switching agents gets cheap because your context moves with you.





