6 karma · joined September 3, 2026
What I spend my time on:
Approval and policy layers so agents can be trusted with risky actions. That's DashClaw: https://www.dashclaw.io
Running Claude Code, OpenClaw, and Codex as daily drivers and open sourcing the setup. My hooks, skills and rules: https://github.com/ucsandman/claude-harness
Turning any API, MCP server or desktop app into a CLI an agent can call: https://declick.dev
An LLM-driven discovery loop that runs overnight. It beat 10 Packomania circle packing records for about $28: https://github.com/ucsandman/discovery-loop
Driving phones and desktops from LLM agents, writing poker solvers (I played pro before any of this), and shipping small products to see what sticks.
Everything else lives at https://www.practicalsystems.io and https://github.com/ucsandman
Based in Ohio. If you're building agent infra, email is wes@practicalsystems.io.
neat idea
Then I had it make a marketing video as a test, pretty neat. Obviously all of this was doable but quite impressed with the model underneath since it's all free, for now.
private data + untrusted content + outbound capability = hard stop
a markdown file is a wishlist, real security lies within hooks and plugins
honestly this was in June which feels like forever ago and I'm not entirely sure what my plan was haha. I knew I was going to be doing marketing for my projects and thought an editor with LLM capabilities would make it easier, but then I just made a repo and a skill where agents can make marketing material directly
I had a similar idea for an opensource video editor where you edit along side the LLM but I gave up on it haha
long live opensource