FIELD NOTE / 2026.09.112 MIN READ / 3 SOURCES

Qingyun Wu, Chi Wang, and AutoGen: Multi-Agent Conversation as a Framework

How AutoGen turned multi-agent conversation into a flexible programming framework for LLM applications in 2023.

AutoGen became one of the defining frameworks of the first LLM-agent wave. Microsoft Research announced it publicly in September 2023 as a way to build complex workflows through conversations among customizable agents.[1] The framework allowed agents to combine language models, tools, human input and different conversation patterns.

Qingyun Wu and Chi Wang are the lead authors associated with the AutoGen research paper, alongside a larger Microsoft and academic team.[2] Their work helped turn “multiple chatbots talking” into a programmable orchestration model.

Conversation becomes architecture

AutoGen’s central move was to treat conversation patterns as a way to compose application behavior. Different agents could hold different roles, use different tools and hand work to one another. The conversation itself became part of the program structure.

That approach fit naturally with LLMs because natural language is both their interface and their internal medium of coordination.

Human participation remains configurable

One of AutoGen’s important design choices was not to assume full autonomy. Agents could incorporate human input alongside model and tool behavior.[2] This made the framework useful for experimenting with different levels of delegation rather than forcing a single definition of an “agent.”

The later AutoGen Studio project further lowered the barrier by providing a low-code interface for configuring multi-agent workflows.[3]

Why AutoGen is a landmark

AutoGen did not invent multi-agent systems, and it was not the only LLM-agent framework to appear in 2023. Its historical importance comes from making agent composition a reusable, open-source programming framework at exactly the moment developers were beginning to explore LLMs as autonomous collaborators.

That combination of timing, flexibility and open tooling made AutoGen one of the clearest bridges between academic multi-agent ideas and the contemporary agent-development ecosystem.

RESEARCH / PROVENANCE

Works Cited

3 SOURCES
  1. 01
  2. 02
  3. 03

CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.

Contribute / Corrections

Improve the record.

Use this moderated submission form to suggest a correction, provide a source, challenge a priority claim or identify a missing contributor. Submissions are treated as research leads, not automatically published comments.

Submit a research lead

Please do not submit confidential material or claims you cannot support.