What it is
A content-monetization protocol on Solana: creators gate content, fans stake tokens for access. I joined the founding week and stayed until my sabbatical, nearly four years later.
What I actually did
Not the on-chain Rust. Another engineer owned that; I worked with it in a passive capacity, building everything around it. My ground: the web platform and its Go services, the design system and shared-package monorepo that let us compose new apps quickly, the content/CMS layer, the embeddable monetization widgets and payment SDKs (public), and the realtime infrastructure I built solo. Toward the end, the AI work: our conversational agent (my ground-up rebuild of an Eliza fork whose code needed serious cleaning), the RAG backend in Python with the team, and an MCP server in March 2025, back when almost nobody had adopted MCP. I published the HuggingFace tokenizers bindings for Node the platform needed as open source.
The honest part
The pivot to AI agents was hype-driven, and it never worked. The agents mostly died; a couple dozen still run in production, flat. But it’s where I learned what agents actually need: memory, sandboxes, isolation, a protocol to reach them. Everything I build now (LoomBrain, sandcaster) descends from what that failure taught me.
What I learned
Watching a company pivot from inside beats reading about pivots. Early standards bets (MCP) pay asymmetrically. And infrastructure conviction survives product failure: the agents died, the thesis didn’t.
The full anatomy of the pivot is in I built the AI pivot from the inside. It didn’t work.