life_agent

Close the gap between information and implementation. For every family. One by one.

The doctor said something; nobody followed up. The bank sent a notice; nobody understood it. The school had a deadline; nobody caught it. That isn't a knowledge problem. It's an implementation gap, and in India it lives in the family WhatsApp group. So the agent wouldn't be an app. It would be the +1 in the group the family already has.

Trust would come in phases: weeks of only watching, then only reminders about time ("the school fee is due in 3 days"), then information but never recommendations, and only then doing things. The agent cannot say the wrong thing if it doesn't speak.

In April and May 2026 I wrote the thesis, a spec and a roadmap, and built a few pieces: onboarding with consent from each member, and a cadence gate with quiet hours and daily limits on interruptions, as fixed rules, not a model's call.

The next step was a 28-day Wizard of Oz pilot, a person playing the agent by hand for 80 minutes a day. It never started. The roadmap had already said the most expensive thing we could do was keep writing.