Arnold

The problem

My health information lived across plans, wearable data, notes, and my calendar. Each source was useful, but none could interpret what was happening in the context of the others or help me decide what to do next.

The approach

I built an agent around the decisions I repeatedly needed to make: how to train, eat, recover, and adjust when real life diverged from the plan. The aim was not another dashboard. It was a system that could remember the plan, understand the day, and help me act.

How it worked

Arnold combines stable personal context with current plans, health data, calendar information, and a daily record of what actually happened. That gives it enough context to make recommendations, notice changes, and explain why an adjustment may be sensible.

Key decisions

• Build around recurring decisions, rather than every possible health question.
• Keep the plan and the record of what happened separate, so the agent can reason about the gap.
• Use the agent as support for judgment, not as a substitute for medical advice.

Outcome

Arnold remains in active use and has led to material changes in how I manage my health. I have used it to rewrite my diet and training plan as my needs changed. Its flags also prompted me to seek physiotherapy for a neck problem that proved considerably more serious than I had initially understood.

What I learned

The system became useful when it gained enough context to understand deviations, not just targets. A technically capable agent still earns trust one recommendation at a time, and specific corrections are more valuable than trying to design a perfect system upfront.