First Person Labs is building Second, a personal-AI product designed to keep an up-to-date model of a user and make it available to different AI agents. Andreessen Horowitz says it is investing about $10 million in the startup, whose approach puts that personal context primarily on the user’s device by default.
Watch Desk analysis
What happened
Second is intended to turn activity on a user’s operating system into a “context vault” that can be inspected and carried across devices and AI agents. First Person Labs says users will be able to choose which intelligence powers it and control access to their data.
The company was founded by Ken Liu and Erik Chi. In its investment announcement, a16z describes the founders as Stanford AI Lab PhD students with backgrounds in AI data privacy and network security. Read a16z’s announcement.
Why it matters
Personal assistants are only as useful as the context they can draw on. Second’s pitch is that this context need not be rebuilt for every new assistant, or locked inside one provider’s service. Keeping the model primarily on-device by default also makes data control a central part of the product’s design, rather than a footnote added after the agents arrive.
That is an ambitious proposition, not proof that the system already works as promised. The announcement does not establish how well Second builds its model of a user, what operating-system activity it can access, or how its privacy controls perform in practice.
Our read
The interesting idea is a portable layer of personal context that users can inspect and share selectively, rather than another assistant asking to know everything and offering little way to take it elsewhere. The test will be whether people get useful continuity between agents without having to surrender more data or wrestle with a second job in settings management.
What to watch
- What operating-system activity Second can use, and how users can inspect or remove it.
- Which devices and AI agents it supports at launch.
- Whether users can move their context easily, and how the company handles security and access controls.
Discussion spark: Would you trust a portable personal context model that stays mainly on your device, or is the promise of better AI assistance not worth building a detailed model of your activity?
Sources and evidence
- Investing in First Person Labs (17 November 2026, 15:10 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.