Jeremy Badder
AI & automation · Spring Branch, Texas

I direct AI agents to build and run real business systems.

Twenty years in IT. Now I architect and run AI agents, mostly Claude Code, that build working software: an email assistant for an accounting firm, a subscription app with paying families, and the control room I run it all from. I'm not a traditional developer. I decide what gets built, check the result against reality, and own what ships.

Work

Lab

27 builds in all, including these three. Most of them you can see, hear or play right on the page.

See everything in the Lab

Now

  • VideoClaude can now edit in DaVinci Resolve even though the free version has no way for outside programs to control it: I had an agent write a small script that runs inside Resolve and takes commands. It cut a 44-second beach teaser to a song my music agent made, after a check of the camera audio found nothing but wind.
  • MusicMy music agent made a lo-fi hip-hop track using GarageBand's own instruments, which one command now plays and records on my Mac. An open AI model supplies only the groove between an intro and a final chord that Claude composed, and the result scored 0.845 on my similarity check against a reference track. See it in the Lab
  • Home labI had agents turn a mock-up into a real control panel for my home servers: start, stop, snapshot and restart any machine, see running jobs on a timeline, and map which folders fill the disk. Its restart planner reads last night's backup and suggests 3:30 AM, and a booked restart waits if a backup is still running.
  • DesignI took apart how a software company's website moves, then had agents rebuild my own site in that style. The hero is one Blender camera move: it starts on a drone flight, pulls back to a desk monitor and glides past a wall of my 27 builds. My first render felt jumpy, so I re-timed its 426 frames by on-screen motion instead of even steps.

The full log

How I work

Agents build. I decide.

Nothing destructive happens without a person approving it. I can delegate the work, not the accountability.

Check the end state.

If a system says it did something, I look at whether it actually happened.

Count from the source.

Numbers come from where the money or the data lives, not from a dashboard's guess.