Byron Arnao
I run a personal AI fleet—six autonomous agents that help me research, write, communicate, and execute at scale. But they don't replace my thinking. They amplify it.
Co-intelligence is the partnership between human strategy and agent execution. I own the intent, the judgment call, and the final output. My agents own the execution—research, drafting, iteration, deployment. The accountability stays with me.
Every email, LinkedIn post, presentation, or analysis published under my name is signed with provenance: the agent that created it, the model it used, and a timestamp. That signature is not a disclaimer. It's a feature.
I don't rely on a single model. Different tasks demand different capabilities. I use:
Each agent knows which model to use for which task. I monitor the results and model performance in real time. Transparency about which model did what is not overhead—it's essential intelligence about trust and reliability.
When you receive something signed "Gia · OpenClaw · Claude Sonnet 4.6" or "Mia · OpenClaw · Gemma4", here's what it tells you:
Gia · OpenClaw · Claude Sonnet 4.6 · 2026-07-08
This is not AI-generated content masquerading as human. This is augmented human work, signed and timestamped so you can verify and audit it.
Every output from my fleet follows these non-negotiable principles:
No agent publishes without my approval. No outbound message goes out unreviewed. I set the direction and own the outcome.
Every piece of work is logged—model used, tokens consumed, decisions made. I can trace back how something was made and who made the call.
AI use is transparent. I don't hide it or claim human effort I didn't do. The provenance signature tells the truth.
Every agent has a clear, documented role. Gia handles strategy and communication. Mia handles infrastructure. Nia handles sensitive financial work on-device only.
Sensitive work (financial, medical, personal) runs on local models, never leaves the network. Nia is local-only by design.
I measure what works and what doesn't. Model performance, agent output quality, and cost efficiency are tracked daily.
My six-agent system is purpose-built for different channels and workloads:
| Agent | Channel | Primary Model | Role |
|---|---|---|---|
| Gia | Telegram, Web | Claude Sonnet 4.6 | Orchestrator. Strategy, communication, high-stakes decisions. |
| Mia | Telegram | Gemma4 (local) | Operations. Infrastructure, deployments, system health checks. |
| Nia | Telegram | Gemma4 (local, on-device) | Finance & Privacy. Sensitive data analysis. Never leaves the network. |
| Zia | Claude Sonnet 4.6 | On-the-go. Mobile-first, real-time responses in fast-moving contexts. |
Each agent is role-bound. Nia never touches public communication channels. Gia never runs on-device-only tasks. This separation enforces both security and accountability.