Overview
Built Fleet Command, a multi-agent AI operating system for running a one-person services business. Specialized agents work across sales, product, delivery, and operations while a Chief of Staff agent manages handoffs, surfaces decisions, and keeps the human operator in control of consequential actions.
Problem
Most agent demos stop at task completion and do not model how work moves across a business. Giving AI too much autonomy creates risk around outreach, scope, claims, and delivery, while fragmented state makes multi-agent work difficult to monitor.
Product approach
The product combines agent orchestration, persistent state, structured telemetry, human-in-the-loop approvals, deployment workflows, and a voice interface. Deterministic gates govern probabilistic AI decisions, including lead scoring, evidence verification, outreach approval, scope validation, and pre-deployment review.
Architecture and workflow
Sidious serves as Chief of Staff. Obi-Wan researches and qualifies opportunities before handing contact-ready prospects to Boba Fett for outreach. Tarkin produces requirements, Vader builds and deploys deliverables, and Leia supports growth and communications. An append-only event architecture records actions, state changes, interventions, runtime, model use, failures, and estimated cost. Git-based delivery, review, reporting, and Netlify deployment are part of the handoff process.
Key lessons
The system supports an end-to-end workflow from discovery and qualification through delivery and retrospective learning. Human control, observable execution, and persistent memory are designed into the operating model rather than added after the fact.
Stack
Claude Code · Anthropic Claude · Gemini API · MCP · Gmail · GitHub · Netlify · VoiceBox · JSONL event telemetry · HTML/CSS/JavaScript · Markdown-based state and memory
