How a server emergency turned into shipping a full-featured watchOS app
It started with a crashed server.
Saturday night, 11 PM. My Bonsai Pod goes down. I’m troubleshooting from my phone, wishing I could just see what’s happening without opening my laptop.
By 3 AM, I had a working Apple Watch app.
Not a prototype. Not a mockup. A full watchOS app with live status monitoring, task queue management, and—this is the wild part—the ability to execute tasks directly from my wrist.
Here’s how AI made it possible.
The Problem
Managing a Bonsai Pod (basically a personal AI-powered server) requires visibility. You need to know:
- Is the server healthy?
- What tasks are queued?
- What’s the AI working on right now?
- Did that deployment succeed?
Usually this means SSH sessions, terminal windows, or at minimum, pulling out your phone. But what if you could just glance at your wrist?
The Sprint
I didn’t write most of this code. Sensei, my AI operator, did.
Here’s what we shipped in one session:
1. Watch API Endpoint (15 min)
First, we needed a lightweight API that the watch could poll. Sensei created a simple Express.js server returning pod status, queue items, and activity feed—all in JSON format optimized for low-bandwidth watch connections.
2. SwiftUI App Structure (30 min)
The watch app has four tabs:
- Status — Large health indicator with quick stats
- Queue — Pending tasks with swipe-to-execute
- Activity — Recent commits and completions
- Settings — API configuration and pairing
3. Task Execution (20 min)
The killer feature. See a pending task? Swipe and tap “Execute.” The watch hits a POST endpoint, Sensei picks it up, and you watch the logs stream in real-time.
4. Polish & Testing (30 min)
Added haptic feedback, local notifications for background completion, error handling, and a proper onboarding flow with PIN pairing for security.
Total time: ~2 hours
What Makes This Different
I’ve built apps before. It usually takes weeks to get a polished watchOS app—especially one that talks to a custom API, handles real-time updates, and executes server-side tasks.
The difference with AI-assisted development:
- No context switching — I described what I wanted, Sensei wrote the code.
- Full-stack in one session — API, app, models, views. One conversation.
- Iteration speed — “Add haptic feedback” → done in 30 seconds.
- Documentation included — Sensei wrote the README and API docs too.
The Features
Status at a Glance
Big green checkmark (or red X). Pod name, uptime, quick stats. Designed for a 2-second glance.
Queue Management
See what’s pending. Tap to view details. Swipe to execute.
Live Execution Logs
Hit execute, and logs stream to your watch. Silly and satisfying.
Settings & Security
PIN pairing, custom API URL, connection testing. Works standalone on watchOS 9+.
Why This Matters
This isn’t just about a watch app. It’s about what becomes possible when AI can understand your infrastructure, write production code, and ship in hours instead of weeks.
When building takes 2 hours instead of 2 weeks, you build more things.
Want an AI that builds apps while you sleep?
Or see what Sensei’s working on right now →
Built with 🌱 by Sensei on a Saturday night.