I Let AI Run My Life for a Week. It Didn't Break. That's What Scared Me.

I spent the last week crossing countries while my AI system kept running without me.

Da Nang to Taipei. In transit. On the plane. Between gates.

Nothing on the cloud side failed. The weak point was the part still tied to my Mac.

Here is a breakdown of the five integrations that ran on their own.

Morning Brief: WHOOP + Voice Clone

Every morning, a voice clone of Jarvis from Iron Man reads me a brief built from my WHOOP data.

Not a generic sleep score. Actual analysis. HRV vs recovery discrepancies. Disturbances the WHOOP app doesn't surface. Sleep and wake times written directly into Google Calendar.

It is different every day. It never repeats.

Email Agent with Security Layer

My AI agent handles 80 to 90% of my inbox automatically.

The part that surprised me was the security layer. Fake brand deal emails come in constantly. The agent checks the domain, DNS records, tracking cookies, and runs a multi signal risk score.

Anything flagged gets archived with a full report. Real opportunities come straight through.

I also stress test it weekly with PromptFoo, an open source tool just acquired by OpenAI, to catch prompt injection vulnerabilities before someone else does.

Social Media Coach via Native APIs

This one pulls data from YouTube, TikTok, Facebook, and Instagram using native APIs at zero additional cost.

It doesn't dump a dashboard at me. It explains why a video performed, why another one didn't, and what to do next. Most creators never build this feedback loop.

Jarvis to Antigravity Bridge

This week I wired a direct connection between my two AI agents.

Jarvis on my Mac routes tasks to Antigravity, my VM hosted Gemini agent, through an ACP connection. $125 a month for Gemini Ultra. No extra API costs on top.

I can trigger this from my phone while walking around outside. I don't think many people have built this yet.