KorvaJoin the waitlist
Tech

An Assistant That Steps In Without Being Asked

It tracked a shifting forecast and revived old conversations, testing whether help should arrive before a request.

AI image
The a16z Show · The Missing Layer for Personal AI | First Person Labs’ Ken Liu

Martin Casado, an early user of an AI system called Second, was planning an outdoor event when the forecast showed a 20% chance of rain.

In real time, the system kept updating the forecast and letting him know whether the group needed to get a tent. He called the help incredibly powerful.

Second, as described by Ken Liu of First Person Labs, is meant to learn about a person from computer activity and send messages when a situation changes.

Fellow early user Joel de la Garza said that kind of help arrives without a request. He said he did not ask for the monitoring, and that he receives texts about an ongoing conversation he had not been thinking about.

He called the messages thoughtful. Casado said the system catches things he would have forgotten.

Liu said the idea is to build a continuing picture of someone and understand what might be important at a given moment. He contrasted that with simpler, reactive help, in which an incoming email triggers another notification.

What he called reflective help connects events over a longer time. It links something said months ago with recent work, a current goal, priorities, a to-do list, and preferences.

For Casado, that meant a changing forecast became practical advice about a tent. For de la Garza, it meant a reminder tied to a conversation that had slipped his mind. Liu's view was plain: "You're not supposed to ask."

Questions people ask

How does an assistant like Second know what I am working on?

It observes signals such as which apps you open, what websites you pause on, what you type, and which networks you join, then summarizes those signals into activities like messaging, research, or writing. The goal is to build a continuing picture of your projects and priorities. You grant permissions for what it can see.

What is the difference between a reactive notification and a reflective one?

A reactive alert tells you something just happened, such as a new email arriving. A reflective message connects events over time, for example noticing that a forecast changed after you discussed an outdoor plan. That second kind requires memory of your goals, conversations, and preferences.

Why don't ordinary phone or computer assistants already do this?

Most assistants only act when prompted, and many run inside one app with limited memory. A persistent assistant keeps context across apps and devices, so it can notice changes without a fresh request. The trade-off is that it must store much more personal information.

More in Tech