Building ProTesla. My own autonomous AI agent.
GPT-4o classifies intent. A router dispatches to 3 specialist agents, each with its own memory, tools, and scope.
Multi-agent. Autonomous. On n8n.
Still building (25% done). Will keep you posted.
#n8n#AIAgents#BuildInPublic
Every business owner has a list of leads they meant to reach out to. It just sits there collecting dust.
So I built a system that works the list for them. Every morning. Automatically.
Here's how it works 🧵
That's the build.
The leads list that used to sit there untouched finally moves, and you still read every message before it goes out.
Recording a full walkthrough soon. Follow if you want to see it.
You set how much it does on its own.
Approve every draft. Approve only the edge cases. Or hand it full autonomy once you trust it.
Tight control on day one. More freedom as it earns it.
The stack is simple:
n8n runs the workflow
Supabase stores the leads
OpenAI does the research and writing
Gmail sends the approved ones
Telegram is the control panel
48 hours after my Claude weekly reset and I've only used 16% of my tokens.
Clearly I'm not pushing my AI experiments hard enough.
I need to start 12 side projects, and put those tokens to work. 😭
An Anthropic engineer showed me why prompting isn't enough.
2:47am. He dropped a video in a Claude dev Discord.
36 minutes long. 290 views.
He explains how to build a loop so your agent can fix itself and finish without you.
Watched it at 3am. By 3:20 my terminal was open. By 3:40 I was already rewriting how I work.
The video disappeared the next day. I kept the mp4. It's below.