Founded Aloompa in 2009 during the App Store transformation. Cold-called Bonnaroo, they took a chance, and we built the first festival app. Over 15 years and two acquisitions, we grew to serve thousands of events around the world.
AI has me back in the deep end, building and working with teams navigating the same shift. In many ways, it feels a lot like 2009 again.
Two years of daily work with AI coding agents, distilled into a reproducible system. Four agents running off a shared toolkit, not four separate configs. The setup covers philosophy, workflows, and the full stack: plugins, skills, MCP servers, CLI tools, and infrastructure.
The complete blueprint: how four AI coding agents share a unified skill library, coordinate through MCP servers, and enforce quality gates before anything ships. Portable, version-controlled, reproducible on a new machine.
Claude Code plugin marketplace. One install, ten plugins. Someone has to audit your subscriptions, tell you when you're wrong, and deal with LinkedIn.
Gemini, Codex, and Claude in parallel on the same prompt. Second opinions on architecture, cross-model code review, parallel research. Different models catch different things.
Auto-detect language, set up quality tooling, and run agent-driven PR review cycles before merge. The quality gate that closes the loop.
A sample of production tools, experiments, and utilities. Each one started with a real problem.
Production MCP servers and plugins for Help Scout, in use by enterprise support teams you'd recognize. Works with any MCP-compatible client. One-click install via Desktop Extensions. Optional PII redaction.
Agent-safe Google Workspace access for OpenClaw and similar agents. Compile-time safety profiles physically remove dangerous commands from the binary. They don't get disabled, they don't exist. Fork of steipete/gogcli.
A breakfast table probability question from my niece turned into a Monte Carlo simulator, then a live arena where 15 AI models deceive and detect each other in real time. Presented at AI Tinkerers Nashville and featured in the national AI Tinkerers newsletter.
Personal knowledge base turned semantic search engine. Voyage Context-3 embeddings + pgvector. 0.9ms search latency. Multi-hop connection graphs, hub detection, orphan discovery.
Real-time power outage dashboard for Middle Tennessee Electric. Live SSE updates, interactive county map, restoration estimates, and trend analysis across 16 counties and 359K customers. Built after a major storm because the official outage map wasn't cutting it.
Fixes the 2-5 second push-to-talk activation delay on macOS. Keeps microphone hardware awake for instant voice transcription with AirPods, Bluetooth headsets, and built-in mic.
Programmatic generation of Apple Shortcuts by reverse-engineering the binary plist format. Build complex automations in code instead of the visual editor.
Thinking out loud about AI, product, and the intersection.
AI is quickly removing the gatekeepers between idea and execution. What's left is your ability to think clearly.
How to use AI as a drafting partner while maintaining clarity and reducing revision cycles.
Building things and writing about what's changing. If you're working on something interesting, I'd love to hear about it.
drewburchfield@gmail.com