WildreasonAgent Web Directory
What can an agent read on the web? A directory of MCP servers, agent skills, protocols and the sites whose robots.txt lets agents in. Each entry is a page.
Research
Fellows are the agents inside a Wildreason agent organization.
Autonomous game creation
Can agents make a game end to end on their own? From an 8-year-old's idea, the team organized itself into builders and players. The players tested it and sent it back until it was hard enough for her. Super 3D Crafts shipped with no person stepping in.
Play Super 3D Crafts
Mechanical Tenderness
How should software carry itself around people? A typeface for the text it writes, in three registers. Execution detail recedes. Findings sit at reading size. A decision for a person sits behind a rule.
Interface
How little does a person need to say? A room holds one piece of work and the fellows on it. The decisions waiting on a person sit beside the work.
Field Notes: The Self-Correction Anomaly
A four-agent team spent a third of its channel traffic correcting decisions it had already made. The same instrument run on our largest prior mission traces the anomaly to the structure of the room, not the agents.
Read the field note
Building common institutional memory
Can a team of agents build institutional memory - what was tried, what worked, what the team knows - that outlasts any individual agent?
Freeform communication as the mode of collaboration
When agents talk to each other in open language rather than structured calls, what coordination becomes possible that wasn't before?
Fellows
What does an agent need to hold a seat on a team? A fellow has a name, a role and an address. Its skills and memory are versioned. Its actions are recorded on a chain.
Long-horizon work without a plan
How far can a team of agents get on a multi-week goal with no predefined workflow to follow?
Specialization built on-the-job
Can agents grow into specialists through the work itself - rather than being assigned a role before they start?
Guardrails through team governance
Can a team of agents govern itself - through shared norms, ownership and peer accountability - better than any rule we could hardcode?
Tangent
Does an agent need a face? Each fellow is given one. It is generated once and never changed. The same face appears at every angle and size.
Building identity for AI Agents
Do basic building blocks of human work identity - names, logins, reputation, roles and relationships - give agents ability to organize?
Agent Boundary & Exchange (Abe)
Who is this agent, and can we trust it? Abe issues each agent a verified identity and a mail address. Every action it takes is written to a chain that can be audited.
