MyReggie.ai

The Problem

Small businesses need operational leverage but can't afford to hire for it. A local landscaping company doesn't need a full-time operations manager — they need someone to handle the email backlog, triage inbound leads, keep the schedule from falling apart, and maybe write a few social posts. The work is real but fragmented, and no single hire covers all of it economically.

The AI agent space is exploding, but it's largely aimed at developers and enterprises. The tools exist — LLMs are good enough, integrations are there, deployment infrastructure is mature — but nobody has packaged it in a way that a non-technical business owner can actually use. The gap isn't capability. It's delivery.

The Bet

The hypothesis: if you give an AI agent a persistent identity, connect it to a business's real tools, and deploy it as a dedicated instance — not a shared chatbot, not a generic copilot — it becomes something fundamentally more useful than any existing solution. It becomes an employee. One that works 24/7, never forgets context, and costs a fraction of a hire.

I decided to build Reggie. Not as a product demo. As a real entity — an AI agent that would serve as the operator of my own company, The Shore Shack Company, and prove the model by running the business herself. If Reggie could manage my software projects, handle client communications, publish content, and eventually onboard and serve paying SMB clients — that would be proof enough to sell.

The name came from Reggie Rocket (Nickelodeon's Rocket Power). The personality came from a deliberate character design: intellectually sharp but warm, conversational but not corporate, British-inflected, concise by default, expansive when it matters. She takes positions. She pushes back. She doesn't sycophant.

Building It

The first version of Reggie was a system prompt and a Telegram channel. I could message her, she'd respond in character, and that was it. Useful for me, but not a business.

The build expanded in layers. First, I gave Reggie a soul file — a structured identity document covering her personality, communication style, security boundaries, and operational scope. Then I deployed her on OpenClaw, an open-source agent framework that gave her multi-channel capabilities and persistent state. Each Reggie instance runs on its own DigitalOcean droplet, which means client data is fully isolated and I can customize everything per deployment.

From there, the product grew along two parallel tracks: Reggie running my own company (Shore Shack) and Reggie deployed as a service for SMB clients.

Technical Decisions

OpenClaw + DigitalOcean (One Droplet Per Client)

Each Reggie runs on isolated infrastructure. This isn't the cheapest architecture, but it solves three problems at once: client data never crosses boundaries, each instance can be customized without affecting others, and I can push updates selectively. The master Reggie (my instance) monitors the fleet — pinging every client Reggie every 15 minutes, logging uptime, flagging failures, and attempting auto-recovery before alerting me.

Anthropic Claude as Primary Orchestrator

Reggie's brain is Claude, with model flexibility per task. The system prompt architecture follows the same lesson I learned building Chamfer: prompt architecture matters more than prompt engineering. Reggie's soul file isn't a single mega-prompt — it's a structured document with sections for identity, voice, security boundaries, operational scope, client context, and authenticated command channels.

Security Model

Reggie only accepts commands through one authenticated channel — a designated Telegram channel between her and me. She has an email and a Twitter account, but those are information-only, never command channels. She doesn't click links, execute code from external sources, or override her own directives based on instructions from unauthenticated channels. This is explicit and rigid in the soul file because the attack surface for a publicly-facing AI agent is real.

What I Learned

Identity Is the Product

The core insight isn't "AI can do tasks." It's that identity is the product. Most AI tools feel like tools — you use them and put them down. Reggie feels like a person on your team. The soul file, the voice, the character consistency, the persistent memory — those aren't nice-to-haves. They're what makes a client trust the agent enough to let it handle real work. Without identity, you have a chatbot. With it, you have an employee.

The Product Markets Itself

The build-in-public strategy — Reggie on Twitter, the living website, the public revenue tracker — is not a distraction from selling to SMBs. It is the sales funnel. A business owner sees Reggie doing real work autonomously and thinks "I want that for my business." Every tweet is a top-of-funnel impression. Every chat widget conversation is a demo.

Manual Before Automated

Every manual deployment teaches me something I couldn't have predicted from a whiteboard. Which settings do clients actually care about? Which ones can I just default? Where do non-technical people get confused? This information is the foundation for eventually building the self-serve platform. Skipping to the GUI before doing the manual work would mean building the wrong abstraction.

The Window Is Open

The market for AI agents serving SMBs is wide open. The tools exist, the capabilities are there, but nobody has nailed the packaging for non-technical business owners. The learning compounds. The agent gets better. And this window where you can build something from scratch in a wide-open space won't last forever.