I've spent two decades inside GMP biologics manufacturing at Merck's West Point site — currently as the SAP PAS-X (MES) subject matter expert, designing master data and electronic batch records that have to pass regulatory audit and survive contact with the actual factory floor.

That's where the governance obsession below comes from. Validation, audit trails, change control, deviation handling — manufacturing spent decades learning how to make automated systems trustworthy. Most AI projects are relearning it the hard way.

A pixel-RPG front-end for real automation stacks — the characters are your actual AI agents, and everything they say is true. The banker knows the balances; the mailman read the email; the approvals desk signs real decisions. Play the demo · Repo

Little people. Real work.

I design and operate reliable AI agent fleets — 47 agents built, 37 in production across finance, rentals, documents, calendar, and email, coordinating to run real operations, not demos. Governed by design: deterministic logic first, AI second, with heartbeats, event logs, and human approval gates on every agent.

flowchart LR
    T([Triggers: schedule, webhook, chat]) --> N[n8n - logic and orchestration]
    N -- writes --> S[(Supabase - source of truth)]
    S -- reads --> L[The Reals - pixel-town interface]
    N -. heartbeats .-> H[workflow_health]
    N -. actions .-> E[agent_events]
    N -- destructive? --> A{approval_requests}
    A -- approved by human --> N
    subgraph Fleet[Agent Fleet]
      C[chief_of_staff]
      R[rental_intelligence]
      F[finance_intelligence]
      D[document_intelligence]
      G[gmail_intelligence]
    end
    C -.coordinates.- R & F & D & G
    N --- Fleet
    classDef brain fill:#06B6D4,stroke:#0891B2,color:#fff
    classDef logic fill:#6366F1,stroke:#4F46E5,color:#fff
    class S brain
    class N,C logic

timeline
    title Good / Better / Best rollout
    Q3 : Rental OS SKU #1
    Q4 : Finance Intelligence rollout
    Q1 : Document Intelligence rollout
    Q2 : Full composable Good/Better/Best menu

1. Discover - map one high-friction operational workflow worth automating.
2. Build - ship one governed, approval-gated agent into production.
3. Prove - measure reliability, speed, and decision quality on real operations.

_{Reliability beats cleverness. Simplicity beats over-engineering.}