I'm really impressed by this use of AI.

Take a photo of a pothole, graffiti, or broken streetlight; AI then analyzes the photo, drafts a report, and submits it to the city.

https://solveto.ca/

#Toronto

With so many AI assistants coming out and getting updated hundreds of times a week, much like with PKM, it's an easy space to get trapped trying to improve your system instead of doing something useful.

If you can build a useful system now, it will only get easier, smarter, and faster over time.

#OpenClaw

I so very much dislike it when applications try to communicate updates through one-time modal windows.

They've done it for years, especially on mobile, and my muscle memory is to dismiss them every time before I ever get a chance to read what they are trying to tell me. Those messages then become lost forever.

I spent the morning experimenting with various popular self-hosted task management and Notion-like applications, but none were very good.

Not terrible applications, but also not any better than what you could do either using plain-text applications like Obsidian with plug-ins, or even Emacs.

I don't want to be fixed into any one tool as things progress so often, but I also don't want to completely change my setup each time a new tool comes out.

This morning, I set up and spent some time using Hermes-Agent. After primarily using #OpenClaw and PicoClaw... It's not bad!

https://hermes-agent.nousresearch.com/

I think this is a pretty interesting use case for MCP.

I want my AI systems to have more access to my Obsidian notes, but I want to be careful that they can't go off the rails unexpectedly with permissions to delete whatever they want

The Task Genius plugin comes bundled with an MCP server, which I can use to let them interact with my tasks, but not my notes as a whole.

Yesterday, I helped a friend with how to approach some problems he was having with a program at work. Today, I'm thinking about judging work using the question, "What comes first?"

If each department only focusing on their own work comes first, then that limits efforts to collaborate. If BRDs come first, then that limits modelling end-to-end flows.

"First" can apply both to sequence and values.

I had an idea yesterday that if I think of my #Obsidian vault as a reflection of what I know (aka my "External brain"), then I could use AI to create a topic map that audits my knowledge gaps.

This can help me identify specific areas to study to fill those gaps. I can then turn this topic map into a plan by either pasting the formatted output back into Obsidian or writing to it directly, complete with tasks and links to notes.