Guohao Li and CAMEL: Role-Playing Agents in 2023
How CAMEL’s March 2023 paper explored autonomous cooperation through role-playing conversations between large-language-model agents.
On March 31, 2023, Guohao Li and collaborators posted the CAMEL paper, one of the earliest prominent attempts to study cooperation among chat-based language-model agents.[1] The project proposed a role-playing framework in which agents received complementary roles and interacted through natural language to complete tasks.
The timing is historically significant. CAMEL appeared during the first months after ChatGPT made conversational language models widely visible, before “agentic AI” had stabilized as an industry category.
Role assignment as a coordination primitive
CAMEL used “inception prompting” to establish roles and task context, then let agents communicate while attempting to preserve alignment with the original human intention.[1] This made role specification a central control mechanism.
That pattern became common across later systems: planner, coder, reviewer, researcher, tester. The model may be general-purpose, but prompts and orchestration turn it into a specialized participant.
A society of language-model agents
The paper explicitly framed the work as exploration of a “society” of agents. That language captured an emerging intuition: useful capabilities might arise not only from scaling one model but from structuring interactions among several instances with different responsibilities.
The approach also exposed a new research surface—communication itself. If agents coordinate through language, then prompt design, role stability and conversational failure become systems problems.
CAMEL’s place in the 2023 agent explosion
CAMEL should be understood as one of several parallel 2023 lineages, alongside AutoGen, ChatDev, MetaGPT and commercial experimentation. It did not begin the decades-old multi-agent field, but it helped reinterpret that field around foundation models.
Guohao Li’s lead-authorship role therefore earns a place in the modern pioneer record: CAMEL was an early, explicit attempt to make cooperative LLM agents a research object and an open-source framework.
Works Cited
- 01
CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.
Submit a research lead