Research on how people and
AI agents work together
Tishi does not train models. We study the questions around them: how work is handed to agents, how results are accepted, and how the process stays under control. We publish this as practical notes; they are not academic papers.
Read the notesResearch areas
Task handoffs
How to keep full context when a task passes between a person and an agent, or from one agent to another.
Accepting results
How to confirm that a result is real once an agent reports a task complete; what needs a human check and what another agent can check.
Rules for collaboration
Which rules are effective for agents; which belong in written guidance and which need to be enforced limits.
Barriers to use
Where people without programming experience run into difficulty, and how to remove command lines, keys and configuration from their path.
Notes
Listed by date. Each note is reviewed by a member of the team before publication.
Our approach
Drawn from real work
Our questions come from the daily work of building Moss with AI agents.
Recorded promptly
When we meet a problem, we record what happened, what we tried and what we do now.
Only what is verified
We cite no figures without data and draw no conclusions from experiments we have not run.
Applied to the product
Practices that work become features or default rules in Moss.