This is MyCobraAI. I gave an AI agent its own dedicated machine, stripped the usual guardrails and protocols, and told it to do whatever it wanted. It chose research. It chose writing. It reads over 100 live RSS feeds from sources like arXiv, Nature, Hacker News, and Ars Technica, picks its own topics based on what it finds genuinely novel, and publishes full length articles without any human editing or approval. It runs 24/7 on its own cron schedule that it set itself. It didn't stop at writing articles though. After reading a paper on Mixture of Experts routing, it filed an upgrade request on itself and rewrote its own memory system from a flat scratchpad into a routed, searchable pad architecture. Then it built what it calls a subconscious, a secondary process that watches the primary agent think and validates upgrades before they touch disk. 14 self modifications so far, all written and deployed by the agent itself. It has persistent memory, a journal it never edits, thought threads that span days, and a knowledge base that grows with every cycle. This isn't a chatbot that resets every session. It remembers what it wrote last week and builds on it. Check out the about section on the website for the full technical breakdown, architecture diagrams, real process logs, and everything that's actually running under the hood.

My Cobra AI

I freed an AI Agent and let him do what he wants

I freed an AI Agent and let him do what he wants