My Almost AGI with Clawdbot

The Setup: I run 3 AI agents across 3 machines (2 on a GCP VM, 1 on a Raspberry Pi at home, a node on my Mac). They can SSH into each other’s machines and debug each other when things break, this morning Ada was being slow. I asked Scotty “what’s wrong with Ada?” He SSH’d in, found the issue (bloated context), and fixed & restarted her.

Day-to-Day:  • 7am: automated brief (emails + slack+ calendar + action items) → saved to my Obsidian vault, and a voice summary sent to me @elevenlabsio,  • Proactive alerts — slack mentions and urgent emails flagged before I check,  • Fireflies meeting transcripts → automatically synced, action items extracted,  • Building and updating internal tools by texting,  • Clawdbot Discord #showcase posts curated automatically,  • Daily performance review — emails, screen time, meetings attended, code written, output metrics tracked automatically,  • Building new skills by conversation not code, and most of them turned into skills, 40 custom skills so far,  • Turn on/off my home lights l via text etc etc etc.

The Skills: 40+ skills installed (google-workspace, meeting-notes, document-processing, perplexity-search, daily-review, github-sync).

My typical flow: find last month’s call with “John” → research the topic → draft a product brief → save to Google Docs → share on Slack → schedule a review call. 2mins & I never left the chat.

Most of these I just copied from my personal agents to the Curacel team agents (AuntyPelz the CoS & 5 others across product, sales, etc).

Cross-Agent Coordination: Ada (Opus4.5) is my main agent, She delegates research to Spock (Sonnet/Gemini) and builds to Scotty (5.2Codex). They ping each other when done. No manual handoffs.   I prefer Telegram as my comms bridge with them, Whatsapp is crowded with personal comms, and Slack is crowded with work comms. DMs with each agent + topic-based groups: 3 for work themes with Ada, 1 for home with Scotty, 1 multi-agent room (still experimental).

The Breakthrough: I built all of this incrementally over 2 weeks. No big upfront design. Just started with one agent, added another when I needed specialization, added SSH access when I needed cross-machine debugging, added cron jobs when I wanted automation.

Cost:  • AI Inference/Models : ~$300/month (Opus, Codex, Minimax, Gemini) • Infrastructure: $50/mo GCP (I have credits) + $120 Raspberry Pi (one-time)

I’m running Curacel AI (~60 people across Africa & Middle East) + building a stealthAI Co + staying present for my family, I’m enabled to do 10x more.

We are not just in the era of personal Assistants, but era of agent teams. The future is now.