Long horizon collaboration

Today's chatbots are designed for discrete tasks like "help me improve this doc". But many things in my life look more like "help me become a better writer" than "help me improve this doc".

If I ask ChatGPT "help me become a better writer", I'll get a plan like this:

ChatGPT's generic writing improvement plan

Even if this were personalized, it still feels like the wrong kind of help. "Help me become a better writer" is not something we can cross off the list right now.

Actually helping me become a better writer means working with me for weeks or months or years. It means understanding things about me that I often can't tell you myself: that my boss thinks I hedge too much, that I lose my voice trying to sound professional, that I tend to move more quickly when I start with an outline. This is not a one-time answer, it comes from working with me over time toward a shared goal.

To explore this, I built a life tracking app with an LLM assistant. (Sidenote: if you're interested in AI products, build your own personal assistant. It's easy and a great playground for ideas.)

It lets me organize conversations around goals: annual, quarterly, monthly. After each interval, I co-author a reflection with the assistant. The assistant also has daily, global, and goal-specific notepads it maintains as its own memory.

There's something motivating about returning to something more stateful than a typical chatbot. While task-specific help is nice, the patterns and questions end up being much more clarifying. As the assistant put it: "I know what you said you wanted at the start of the year. I know what you actually did. I'm here to ask about the gap."

Life tracking app screenshot

Note: it's funny to me Opus 4.5, when given carte blanche, decided to choose Anthropic's color scheme.

A few questions on my mind: