AgentCoder: Multi-Agent-based Code Generation with Iterative Testing and Optimisation

Abstract

We present AgentCoder, a multi-agent framework for code generation that uses separate agents for programming, test design, and test execution. The iterative testing loop significantly improves code quality.

Key Contributions

• →Programmer-tester-executor agent triad
• →Iterative code improvement
• →Test-driven agent development

Topics