What is the right abstraction for a personal AI assistant?
I think with a little effort, I can come up with a better abstraction for using personal AI assistants.
Why is this necessary? Can't I just use Muse or Claude on Mac Mini and not over-engineer things? Here are some thoughts, in no particular order.
- Different personal assistants have different strengths, like Muse using Instagram or Grok using X.
- Free tokens!
- I don't want my workflow and context to be locked in to a single personal assistant.
- There are existing best practices for working with personal assistants that aren't being used.
- Some assistants are IP-blocked / banned from certain websites (eg. Amazon).
- I also want to use the other LLMs (Claude / ChatGPT).
- A Mac Mini won't be IP blocked by certain websites if used sparingly.
- I can be deliberate about the separation between the assistant and my personal accounts (eg. giving my assistant a separate phone number and accounts).
I'm not sure if I want a single assistant that uses other services, or several assistants working concurrently and updating shared state.