Independent research collective
A2A Lab studies the software systems that make AI agents useful, governable, and capable of sustained work.
Our research connects enterprise software practice with controlled studies of agent interfaces, execution, communication, and adaptation.
01 / Publications
Public research from A2A Lab.
Full texts available on arXiv.
A semantic interface for enterprise agents, organizing tool use around search, resolution, preview, execution, verification, and recovery. The study evaluates the approach against conventional interfaces on operational tasks.
An execution framework that adjusts resources and review to task risk, separates proposal from verification, and incorporates recovery when execution fails.
02 / Research
From the information an agent receives to the consequences of its actions.
What must software expose for an agent to work with it?
We study how business objects, current state, evidence, and available actions can support agents in completing and verifying work.
How can autonomy operate within clear authority?
We examine risk-sensitive execution, independent verification, action boundaries, and the records needed to trace responsibility.
How do agents influence one another through a shared environment?
We investigate communication and collaboration through shared resources, and how information moves between agents over time.
How can systems observe and improve their own operation?
We study how agents can support software adaptation and maintenance, and how proposed changes can be evaluated before use.
03 / People
A shared research agenda in agent-native enterprise software, organizational AI governance, multi-agent coordination, accountability tracing, and human–AI collaboration.
Researcher · Agent-native systems
A2A Lab
Professional Member of the China Computer Federation (CCF)
Researcher · Agent-native systems
A2A Lab
04 / Contact
Research collaboration, replication questions, and correspondence.