From RAG Pipeline to AI Agent: 14 Months of Building a Production Chatbot
How our enterprise chatbot evolved from GPT-3.5 with chunks in the system prompt to a full AI agent — and why every RAG trick we built along the way became obsolete.
How our enterprise chatbot evolved from GPT-3.5 with chunks in the system prompt to a full AI agent — and why every RAG trick we built along the way became obsolete.
How I built a novel UI-tree capture pipeline — browser bookmarklet, tree server, and AI generation — that produced 715 structured RAG actions where months of manual work had yielded 3.
How I built a Model Context Protocol server that gave AI agents full codebase understanding — cutting task completion time by 4.7x on a 500k+ LOC legacy system.
What 8 years of architecture decay actually looks like in a 500k-line enterprise monolith, how I diagnosed the root causes, and the four options I presented to the client — one of which was deliberately provocative.
How I built a distributed bot fleet that farmed MMO gold, listed it on eBay, extracted buyer names with GPT-4o, delivered gold in-game, and confirmed fulfillment — all without a single human in the loop.