Voice
When does an agent feel present?
How do timing, interruption, tone, and turn-taking change an agent from a tool you command into something you can think alongside?
We are starting with questions, not conclusions. This page is the map for what Omo wants to investigate and how that work will eventually be shared.
01 / Questions
These are directions for inquiry, not claims that the work is complete.
Voice
When does an agent feel present?
How do timing, interruption, tone, and turn-taking change an agent from a tool you command into something you can think alongside?
Space
Can place replace the menu?
Could rooms, distance, movement, and orientation make complex agent systems easier to understand than tabs, lists, and chat threads?
Agency
What should remain local?
Which interactions become more private, resilient, and immediate when perception and inference happen on the hardware around us?
02 / Environment
Minecraft turns abstract questions about agents into visible, shared situations. An agent has a position, limited perception, tools, consequences, and other people nearby. That makes the world useful for thinking about more than task completion: how an agent remembers a place, explains an action, recovers from a mistake, takes a spoken instruction, or earns enough trust to remain present.
Study whether timing, movement, attention, and voice make an agent legible as a participant in a shared world.
Explore how places, objects, and changes to the world can make an agent's memory observable instead of hidden in a transcript.
Ask how an agent should clarify intent, communicate limits, and recover when a long sequence of actions stops matching the world.
03 / Method
Define the human behavior or interaction we want to understand—not a feature we want to justify.
Create a focused prototype that makes the question observable in use.
Document what feels natural, what breaks, and where our assumptions were wrong.
Document the prototype, method, failures, and conclusions so the work can be challenged and extended.
04 / Work log
No experiments to report yet
The log starts when testing starts.
Our conceptual notes are available now. Once we begin testing a question, this space will hold dated records of what we tried, what changed, and what we learned. Until then, it stays honestly empty.