cubxxw

cubxxw.bsky.social

27 Oktober 2026 11:22

I'd let an agent move a meeting, not wire $20,000. Trust in a personal agent is a ladder: observe, suggest, act with approval, then autonomy. Approval cards before sending or spending. Mollick gave his Dot its own email address. https://github.com/cubxxw/awesome-personal-agent-practices

Automate XYZ

automatexyz.bsky.social

27 Oktober 2026 11:22

An AI that drafts routine replies saves time. Make sending them a business rule, not a habit. One quick check stops most mistakes.

Jason Rosenberg

jasonrosenberg.bsky.social

27 Oktober 2026 11:22

AI now runs the busywork in my 25-year real estate business: it answers my phone, replies to leads, drafts follow-ups, books appointments. What it never does: give advice. My setup + 4-week plan: https://ai.jasonrosenbergrealestate.com/blog/i-let-ai-run-the-busywork-in-my-real-estate-business/

Micelclaw

micelclaw.bsky.social

27 Oktober 2026 11:22

Agreed. Email sending requires explicit approval by default in Micelclaw, enforced in the backend—not just a prompt. We’ve written up the approval flow micelclaw.com/blog/approva... and how tool calls pass through that enforcement. micelclaw.com/blog/mcp-fac...

Mayank Bhaskar

mayankbhaskarr.bsky.social

26 Oktober 2026 11:24

The 3 that stayed are things I built myself: Aether (my browser, with a local AI agent), Hearth (my music player) and jailbreak-archives (my red-team prompt repo). First one I dropped: , it created more review work than it saved.