Systems on Glass · Episode 13 · AI agents

Stop Babysitting

or: allow, ask, deny

The 47-second version. The full video is coming to YouTube; the link goes here when it's up. The narration is an AI clone of my voice.

Made with AI, for fun

Systems on Glass is an AI-automation side project of mine. Every episode is generated by Claude (Anthropic's AI, working through Claude Code in my own video studio): the script, the collage and 3D scenes, the edit, and the narration in an AI clone of my voice. I choose the topics, bring the stories from my own work, and review each episode before it goes up. Not affiliated with Anthropic.

Every time my agent runs a command, someone has to approve it. Approve everything by hand and you stop reading what you approve; say yes to everything and it can spend your money without asking. While making this episode, I found out mine could.

The short version

"allow": [46 free studio commands] · "ask": ["Bash(*--yes*)"] · "deny": ["Read(./studio/.env)", "Read(**/.env)"]
evaluated: deny → ask → allow, first match wins · sandbox: 84% fewer prompts (Anthropic, Oct 2025)

The fix is a real commit in my studio, made the day this episode was recorded.

Further reading

Next episode: Cache the Model. Previously: Idempotency. All episodes →