I built a small AI-powered business operating system using Claude Code and ChatGPT. Am I overengineering this?
retained text
I’ve been working on a startup and over the past few weeks, I’ve been building something I call **Nexus**. The idea was pretty simple: Instead of constantly switching between AI tools, copying conversations, and losing track of projects, I wanted one system where different AI agents could help me run and build a business. Here’s what I’ve built so far: **1. Nexus — The central dashboard** A place to track projects, tasks, business decisions, progress, problems, and what different agents are working on. **2. Bibi — My main AI developer** Built around Claude Code. It handles coding, debugging, building features, testing, and technical implementation. **3. Bob — An AI task executor** Bob takes assigned tasks, uses Claude Code to work on them, tracks progress, handles interruptions, and reports the results. It has pause/resume controls and limits to prevent uncontrolled execution. **4. Sherlock — The quality checker** Instead of trusting an AI that says “done,” Inspector runs checks, verifies file changes, looks for potential credential leaks, and determines whether a task passes. **5. Shared AI memory** I use a private GitHub repository to store business decisions, project status, agent responsibilities, and handoffs so ChatGPT and Claude can work from the same information. It’s not perfect automatic memory. Each assistant still needs to read and update the shared records. **6. AI model router** I built a system for managing free AI API providers, checking availability and restrictions, and potentially assigning suitable tasks to different models. Most integrations are still disabled or unverified. **7. Sales automation** There’s a lead research and qualification system, a CRM, personalized email drafts, and a review process. Currently, it has 132 leads, 101 qualified, and eight complete draft sequences. No prospect outreach has been sent yet. **8. Backups and synchronization** Nexus can create backups, test restoration, and publish a read-only HTML dashboard snapshot to GitHub. This is actually where I ran into problems. A lot of my work still exists only on my main computer, and synchronization isn’t fully automatic. I’m working on fixing that. **Where things stand:** The core systems are built, and Claude has reported over 180 passing tests. But this is still an experimental internal setup, not a fully autonomous business. I haven’t acquired my first paying customer through it yet. My ultimate goal is to have a small, affordable AI-assisted company where agents help with development, research, operations, and repetitive work, while I focus on decisions, sales, and customers. **My questions for you guys:** Am I building something genuinely useful, or have I overengineered what could’ve been a much simpler workflow? What would you improve, remove, or approach differently? I’d appreciate criticism, especially from people who’ve built multi-agent systems or used Claude Code extensively.
document fetched 2026-10-30 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xi7y5z/i_built_a_small_aipowered_business_operating