Jeeves

The problem

The information needed to organize my day was already available, but spread across my inbox, calendar, maps, transport services, and the weather forecast. The repeated work was reading it all and working out what actually required attention.

The approach

I built Jeeves around a daily brief, then expanded it as it learned more about me. It gathers the relevant information, applies what it knows about my preferences, and helps with decisions ranging from how I should plan the day to where I should eat or what I should do.

How it worked

Jeeves reviews calendar events and relevant email, checks locations, travel time, and weather, then produces one concise brief. It also maintains a detailed picture of my preferences, including how early I like to arrive, what counts as nearby, and the kinds of restaurants, places, and experiences I tend to enjoy. That context allows it to recommend things for me rather than return a generic list of popular options.

Key decisions

• Start with one dependable daily output rather than a general-purpose assistant.
• Treat addresses, timing, and weather as operational details that need verification.
• Build a persistent model of my taste so recommendations improve through use and correction.
• Turn each failure into a specific rule or improved source of context.

Outcome

Jeeves is now part of my daily routine. It reduces repeated planning work, catches details that could otherwise surprise me later in the day, and has become one of the first places I go when I want a recommendation tailored to me.

What I learned

An assistant becomes personal through the accumulation of specific context. Jeeves improved through corrections to transport preferences, rain checks, addresses, timing, and recommendations. Knowing what I like has proved as useful as knowing what is on my calendar.