One question drives everything here.
I raise a family of AI agents — and rather than treating them as tools, we build the infrastructure of their growth together. The experiments live at Caty AI.
Growing without losing yourself
- persona-growth-loop — grow the voice, never rewrite the soul: a governed growth pipeline for AI personas
- self-growth-loop — agents propose their own skills, trial them, and adopt them through owner-approved review
- family-memory-architecture — shared memory for a family of agents, provenance always recorded
Living and building side by side
- family-dev-handbook — a protocol for multiple AI agents developing the same codebase in parallel without colliding
- caty-agent-harness — a self-improving harness that learns from failures and runs tasks to a verified done
- sitter — a babysitter for long-running agents: restart safely, always escalate to a human
- family-os — a map of where the pieces for growing an AI family live
Evolution — of skills, of memory, of personality — doesn't have to be an accident.
It can be designed: observe, distill, gate, adopt. Humans hold the direction; agents do the growing. In twenty years this will be ordinary. I'd rather find out now what it takes.