AGILECODER is a novel multi-agent software development framework inspired by Agile Methodology (AM). It integrates specific AM roles—Product Manager, Developer, and Tester—into different agents, enabling collaborative software development based on user inputs. The system enhances development efficiency by organizing work into sprints and focusing on incremental development through sprints. Additionally, it introduces the Dynamic Code Graph Generator (DCGG), which creates a Code Dependency Graph (CDG) that dynamically updates as the codebase evolves. This allows agents to better understand the codebase, leading to more precise code generation and correction. The framework evaluates its performance on established benchmarks like HumanEval and MBPP, as well as a new benchmark, ProjectDev, which includes more complex software development tasks. Experimental results show that AGILECODER achieves state-of-the-art performance, outperforming recent SOTA models such as MetaGPT and ChatDev. The framework's success highlights the potential of integrating Agile Methodology and static analysis techniques into multi-agent software development. Key contributions of AGILECODER include: 1. **Multi-Agent Framework**: Aims to mimic the AM workflow in a multi-agent context, emphasizing effective communication and incremental development. 2. **Dynamic Code Graph Generator (DCGG)**: Creates and updates a CDG to capture evolving code relationships, enhancing context-aware code retrieval. 3. **State-of-the-Art Performance**: Achieves superior performance on benchmarks like HumanEval, MBPP, and ProjectDev, outperforming leading models. The paper also discusses limitations and future work, including the potential for integrating additional Agile practices and extending the framework to other domains. Overall, AGILECODER represents a significant advancement in the automation of software development using multi-agent systems and Agile Methodology.AGILECODER is a novel multi-agent software development framework inspired by Agile Methodology (AM). It integrates specific AM roles—Product Manager, Developer, and Tester—into different agents, enabling collaborative software development based on user inputs. The system enhances development efficiency by organizing work into sprints and focusing on incremental development through sprints. Additionally, it introduces the Dynamic Code Graph Generator (DCGG), which creates a Code Dependency Graph (CDG) that dynamically updates as the codebase evolves. This allows agents to better understand the codebase, leading to more precise code generation and correction. The framework evaluates its performance on established benchmarks like HumanEval and MBPP, as well as a new benchmark, ProjectDev, which includes more complex software development tasks. Experimental results show that AGILECODER achieves state-of-the-art performance, outperforming recent SOTA models such as MetaGPT and ChatDev. The framework's success highlights the potential of integrating Agile Methodology and static analysis techniques into multi-agent software development. Key contributions of AGILECODER include: 1. **Multi-Agent Framework**: Aims to mimic the AM workflow in a multi-agent context, emphasizing effective communication and incremental development. 2. **Dynamic Code Graph Generator (DCGG)**: Creates and updates a CDG to capture evolving code relationships, enhancing context-aware code retrieval. 3. **State-of-the-Art Performance**: Achieves superior performance on benchmarks like HumanEval, MBPP, and ProjectDev, outperforming leading models. The paper also discusses limitations and future work, including the potential for integrating additional Agile practices and extending the framework to other domains. Overall, AGILECODER represents a significant advancement in the automation of software development using multi-agent systems and Agile Methodology.