← Work

An agent older people can talk to by text

People want help keeping their day on track and staying connected with family, without learning a new app.

Indy logoIndy · Co-Founder and Sole Engineer
indy.day2026 – present

The problem

Indy is an AI-native agent that helps people keep living life their way. You talk to it by text, so there's no new app to learn. It remembers things, connects to calendar and email, sends a morning rundown of what's ahead, and organises plans with family, friends and the people who support you.

Constraints

  • People trust it with their day. How it behaves matters as much as what it can do.
  • Conversations have no fixed shape. Someone mentions a fall in the middle of a message about lunch, or says "send that to my daughter" without saying which message or which daughter.
  • Real actions. Indy sends email and changes calendars, so a mistake isn't just a bad reply.
  • Models and instructions keep changing. Any change can quietly break a behaviour that used to work.

What I did

I co-founded 388 Labs, the company behind Indy, and build Indy as the sole engineer. The product rests on a few decisions:

  • Text first. Indy lives at one phone number that people save as a contact. There's nothing new to install or learn, and their mobile number identifies them.
  • Indy shows up. A daily check-in brings together what's ahead, so nobody has to remember to use it. Indy follows up on plans and closes the loop when they're done.
  • Memory the person controls. Indy remembers what it's told, from conversations, calendar and email, and the person can correct or change anything it knows.
  • The people around them. Family, friends and support people can share plans with Indy, so the person doesn't have to pass them on, and Indy coordinates changes with them.

A large part of the work was making Indy's behaviour testable. Every piece of behavioural guidance comes with eval scenarios that check it, such as spotting distress, spotting scams and staying out of medical advice, and the full suite runs on every change to guidance, tools or models. Where a boundary is critical, it's enforced in code as well: Indy has no tool that can share bank details or make a payment. I wrote about the approach in How to test whether your AI agent follows instructions.

Outcome

Indy is live at indy.day, with people using it.