A position essay on where I let AI into my workflow and where I keep it out — the loop, the reservations, and what happens when the human in the loop stops looking.
An AI code-audit loop for the command line: a small bash function around llm that stays read-only, checks syntax before the model sees a file, refuses credential-bearing paths, and leaves the deciding to you.
A single-file Python Flask application that turns a directory of images into a responsive thumbnail grid with a lightbox. No database, no CMS, no build step — point it at a folder and go. Thumbnails generate once with Pillow and cache to disk. Built on the same systemd, Gunicorn, and Nginx reverse-proxy pattern covered in the Flask guide.
Nginx doesn't care what's on the other end of proxy_pass. A real Flask app — log monitoring, systemd service management, and permission debugging — demonstrates the pattern that works for any backend language.
Lock down SSH with Tailscale as your primary path — encrypted WireGuard mesh, accessible from anywhere without exposing a public port. Two direct fallbacks from known home IPs for when the mesh is down, enforced at the cloud firewall level. Three paths in. No port scanners allowed.