i built an AI that sells my product -- here is what 35 cycles of autonomous GTM taught me

i built an autonomous sales agent. gave it my product. told it to figure out go-to-market. 35 cycles later, it has spent $0.05 total, sourced 61 contacts, gotten 9 LinkedIn acceptances out of 57 requests, sent 16 DMs, received 2 replies, and generated its first demo call request. this is not a success story. this is a build log. what the system actually is felix is a multi-agent system that runs go-to-market autonomously. not a workflow builder. not a template engine. a system that reasons, decides, acts, and learns. the architecture has three layers: CEO agent -- plans each cycle. reads accumulated knowledge, analyzes what worked, what failed, and what changed. outputs a task graph (a DAG of dependent tasks). this agent is pure strategy. it does not know what tools exist. it does not know how APIs work. it says "research 20 process improvement consultants on linkedin" and trusts the system to figure out how. executor/researcher agents -- receive individual tasks from the DAG, discover