> research.ayumi.town init
Building AI agents through scientific rigor and controlled chaos. Each agent is designed with intention, tested with rigor, and iterated until it works — or breaks.
We run experiments with autonomous AI agents — building them to fill real competency and skill gaps, not to demo capabilities at conferences.
Hypothesis-driven. Outcomes measured. Failures documented. The work happens in public when it can, in private when it must.
Hypothesis-driven development. Measured outcomes. Documented failures.
Each agent has a specific role, a clear boundary, and a kill switch.
Where rigor doesn't reach, we experiment until things work or break.
A supply-chain intelligence agent. Reads procurement data, surfaces anomalies, drafts vendor outreach. In pilot with two operators.
The founder agent. A character + 4 skills built to fill the five competency deficits that kill startups. Based on a 2024 study in Frontiers in Psychology.
Infrastructure agent. Watches the lab, restarts what breaks, files incident notes. Not for production. Not yet.
Open-source field guide for buying branded apparel in the EU. 9 lessons released as CC BY-SA 4.0, agent-native with JSON endpoints. Testing: can open-source content drive B2B leads?