Let's give it some data

Next up, for my Bot project, I needed to get Claude looking at my calendar. To do this I setup specific credentials with FastMail that have access to my calendar -- read access for now. I installed a CalDAV library for Python, and then, using code Claude wrote, updated my bot.

The first time I needed to paste some code into Claude, I made an architectural decision here that Claude hadn't suggested - I wanted to setup a secrets file so that I could more easily shuttle my bot script back and forth to Claude without needing to strip out keys every time. I knew of this concept from my prior work with Python and, honestly, Home Assistant. However I couldn't remember exactly the best way to do it. So I just asked Claude. 

After some rounds of back-and-forth, we got the setup working! It's pretty amazing to see it working.

What I'm learning so far:

  • The architecture gets complicated fast. I already have multiple services talking to each other (Telegram, FastMail, Claude, and the python script running on my server)
  • There's still a lot of engineering involved. I can sort of squint and see there's a bunch of problems I need to solve still. It's not as easy as "wire 6 services together and call it a day."
  • It makes me wonder how or where I might go get some training on just good architecture patterns and habits without spending the next 2 years doing a CS degree. I know a bit from scripting and writing very simple programs over the years, but I have learned it haphazardly at best. 

I'm heading toward a multi-agentic approach, where I have a bot or thread going for each major area of my life (Here's the Fitness bot, here's the Home Assistant bot, here's the Career Development bot, here's the complete project xyz bot...).

What I need to make some decisions about / am curious about:

  • How do I want to handle chatting with each of these? Or do I keep these in the normal consumer version of Claude and just keep something like a master project summary file up to date?
  • How do I want to handle stuff that doesn't fit cleanly into any of the specialist buckets I'm imagining.

Immediately I'm sensing that I like the idea of having a separate setup for each major goal or project I'm working on AND I can see how that may get very tedious to manage.