What I'm building.
Three projects, one ecosystem. All built after hours by one person. They share infrastructure, accumulated knowledge, and hard-won lessons.
Autonomous Agent Ecosystem
I wanted to stop doing repetitive work manually. So I built 14 AI agent tasks that run on their own — syncing, diagnosing, fixing what they can, escalating what they can't. They share one protocol (Preflight, Execute, Emit), one event log, and one knowledge store. The goal is an ecosystem that gets better without me having to coordinate it.
It's not there yet. But tasks can now detect each other's failures, auto-fix common issues, and propagate what they learn. The three-layer escalation chain handles most routine problems without human intervention.
Relationship-Driven AI Design
What does it take for an AI relationship to feel real over months, not minutes? Not a chatbot with personality settings — something that remembers, that has opinions, that can say no. I've been running this experiment for a while.
The core insight so far: relationship quality is not a function of model capability, but of what you accumulate between sessions. The rest is engineering — and I'm still figuring out how much of it I can share.
Sprint-Based Game Development Loop
I set the direction; AI agents handle implementation, QA, and regression testing. It's a director/implementer model — the human makes design decisions, agents write the code and catch the bugs.
25+ sprints in, rebuilding the first 5-minute experience. Still learning what works and what doesn't when you develop a game this way. The biggest lesson so far: agents are great at execution, but design taste is still a human job.