Hongyi Du

Researcher in AI agents and organizations

UIUCUrbana–Champaign

Building the
organizational
layer of AI.

I study how AI systems evolve from models to agents, from collaboration to organizations, and from human–agent organizations to agent societies.

MY RESEARCH, ACROSS SEVEN SCALESScroll to explore
RESEARCHER / UIUC

Hongyi Du 杜洪一

I work on AI agents, their organizations, and the interfaces that connect them to people. At the University of Illinois Urbana-Champaign, my research spans agentic problem solving, communication protocols, and persistent organizational capability.

01 / MY RESEARCH AGENDA

Intelligence is a beginning.
What comes after?

I explore these questions at different scales, through the projects and papers connected to each stage.

SHARED NORMS · INTEROPERABILITYCROSS-ORGANIZATION INSTITUTIONSCOOPERATING ORGANIZATIONSINTER-ORGANIZATION NETWORKSHARED INFRASTRUCTURE & COMMONSRESOURCESPUBLIC EVIDENCESHARED INTERFACESONE TASK · COMPLEMENTARY WORKSHARED DELIVERABLEContributions → integration → shared outcomeINTERFACES · ROUTING · MESSAGE EXCHANGEProtocolRouterA2AACPANPAgoraRequests ↔ responses · protocol-dependent pathsA RUNNING SENSE–ACT–ADAPT LOOPOBSERVEACTFEEDBACKPAST EXPERIENCECURRENT POLICYNEXT ADAPTATIONGOVERNANCEORGANIZATION-OWNEDPROPOSEREVIEWADOPTAUTHORREVIEWERINTEGRATORSHARED VALIDATIONEVIDENCE → REVISIONSHARED STATE · WORK RECORDHUMANLIAISONSHARED REVIEWHUMANHUMAN
RESEARCH SCALES / 01Conceptual structure
EPOCH 01Technical starting point

Model

What can a model understand?

A latent intelligence core with capability, but without agency.

CapabilityRepresentationPotential
Continue the story
EPOCH 02Published research

Agent

How does intelligence become action?

The model becomes an acting unit with goals, tools, and interaction with the world.

ModelingAgentFindings of EMNLP 2025
Tool useActionTask execution
Continue the story
EPOCH 03Published research

Multi-Agent

When does collaboration help?

Multiple agents begin to interact, coordinate, and sometimes interfere with one another.

ACL 2025
ICML 2026
CommunicationCoordinationConflict
Continue the story
EPOCH 04Research question

Autonomous System

What keeps an agent system running?

Agent interaction becomes a continuously running system with shared state and runtime structure.

RuntimeShared stateContinuous execution
Continue the story
EPOCH 05Current research focus

Agent Organization

What makes capability organizational?

Capabilities become organizational when they are encoded into persistent structure, responsibility, and governance.

RELICCurrent research
External evaluation · CooperBench / ProgramBench
GovernanceExecutable protocolsInstitutionalization
Continue the story
EPOCH 06Exploratory HCI extension

Hybrid Human–Agent Organization

How should humans enter the loop?

Humans interface with an organization through judgment, authority, and translation layers.

Human–Agent OrganizationExploratory research
LiaisonOversightEscalation
Continue the story
EPOCH 07Long-term direction

Agent Society

What exists beyond one organization?

Beyond a single organization lies a larger ecology of interacting organizations, roles, norms, and institutions.

Agent SocietyFuture direction
InstitutionsRelationsSocial structure
Explore the research map
02 / THE RESEARCH MAP

One agenda.
Different scales of inquiry.

Explore my research
04 / CURRENT QUESTIONS

Questions I’m
exploring.

IN FOCUS / RELIC

What persists when
the members change?

Agent OrganizationCapabilities become organizational when they are encoded into persistent structure, responsibility, and governance. RESPONSIBILITY / GOVERNANCEPLANEXECUTEREVIEWPERSISTENT ORGANIZATIONAL STATEMEMORY / REUSABLE CAPABILITYGATEMEMBERS CHANGE. ORGANIZATIONAL STATE PERSISTS.

Organizational memory, responsibility, and reusable capability across tasks.

Follow the research
05 / RESEARCH NOTES

How I think about
agents working together.

All notes
MultiAgentBenchCONCEPT / RESEARCH SYSTEM
ONE TASK · COMPLEMENTARY WORKSHARED DELIVERABLEContributions → integration → shared outcomeINTERFACES · ROUTING · MESSAGE EXCHANGEProtocolRouterA2AACPANPAgoraRequests ↔ responses · protocol-dependent paths
Complementary contributions · a shared outcome
01
DRAFT / RESEARCH NOTE

Why agent cooperation fails.

When more agents create more coordination work—not better work. From MultiAgentBench to persistent organizational capability.

Agent cooperation4 min read
RELICCONCEPT / RESEARCH SYSTEM
Agent OrganizationCapabilities become organizational when they are encoded into persistent structure, responsibility, and governance. RESPONSIBILITY / GOVERNANCEPLANEXECUTEREVIEWPERSISTENT ORGANIZATIONAL STATEMEMORY / REUSABLE CAPABILITYGATEMEMBERS CHANGE. ORGANIZATIONAL STATE PERSISTS.
Governed protocols · capability beyond members
02
DRAFT / RESEARCH NOTE

AI agents are still in the organizational stone age.

Modern individual capabilities, primitive working arrangements: the missing language, institutions, and continuity of collective work.

Organizations4 min read
ProtocolBenchCONCEPT / RESEARCH SYSTEM
ONE TASK · COMPLEMENTARY WORKSHARED DELIVERABLEContributions → integration → shared outcomeINTERFACES · ROUTING · MESSAGE EXCHANGEProtocolRouterA2AACPANPAgoraRequests ↔ responses · protocol-dependent paths
Protocol interfaces · routing · message exchange
03
DRAFT / RESEARCH NOTE

Protocols are a foundation for agent society.

From communication protocols to generated, governed rules of work—and the institutions between organizations.

Protocols4 min read
06 / PUBLICATIONS

Research papers.

All publications
2026

ProtocolBench: Which LLM MultiAgent Protocol to Choose?

Hongyi Du, Jiaqi Su, Jisen Li, Lijie Ding, Yingxuan Yang, Peixuan Han, Xiangru Tang, Kunlun Zhu, Jiaxuan You

ICML 2026First author
2026

Ψ-Bench: Evaluating Persona-Sensitive Influencing in Persuasive Dialogues

Peixuan Han, Hongyi Du, Jiayu Liu, Yihang Sun, Jiaxuan You

Preprint
2025

MultiAgentBench: Evaluating the Collaboration and Competition of LLM Agents

Kunlun Zhu, Hongyi Du, Zhaochen Hong, Xiaocheng Yang, Shuyi Guo, Zhe Wang, Zhenhailong Wang, Cheng Qian, Xiangru Tang, Heng Ji, Jiaxuan You

ACL 2025Core contributor · co-first author
2025

ModelingAgent: Bridging LLMs and Mathematical Modeling for Real-World Challenges

Cheng Qian, Hongyi Du, Hongru Wang, Xiusi Chen, Yuji Zhang, Avirup Sil, Chengxiang Zhai, Kathleen McKeown, Heng Ji

Findings of EMNLP 2025
2025

EscapeBench: Towards Advancing Creative Intelligence of Language Model Agents

Cheng Qian, Peixuan Han, Qinyu Luo, Bingxiang He, Xiusi Chen, Yuji Zhang, Hongyi Du, Jiarui Yao, Xiaocheng Yang, Denghui Zhang, Yunzhu Li, Heng Ji

ACL 2025
RECENT UPDATES

From my research.

All updates
OPEN CONVERSATION / Home

Continue the conversation.

Questions and perspectives on this page are welcome.

Prefer a private conversation?

Loading comments…

Leave a public comment

This conversation belongs to Home. Comments appear only after review. For contact details or personal matters, use the private message form.

Public · reviewed before appearing