A survey of Code Agents for improving development productivity. These agents aim to help
- SWE (Software Engineer)
- MLE (Machine Learning Engineer)
- DS (Data Scientist)
- DA (Data Analyst)
| Paper | Year | Publisher | Type | Institution | Code |
|---|---|---|---|---|---|
| SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering | 2024 | Arxiv | GitHub issue fixing | Princeton |  |
| SWE-bench: Can Language Models Resolve Real-World GitHub Issues? | 2024 | ICLR | benchmark github issue fixing | Princeton |  |
| DevBench: A Comprehensive Benchmark for Software Development | 2024 | Arxiv | benchmark LLM for dev lifecycle | Shanghai AI Laboratory etc |  |
| Code Generation with AlphaCodium: From Prompt Engineering to Flow Engineering | 2024 | Arxiv | code generation for CodeContests | CodiumAI |  |
| ChatDev: Communicative Agents for Software Development | 2023 | Arxiv | design, coding, and testing | Tsinghua University etc |  |
| MetaGPT: The Multi-Agent Framework | 2023 | Arxiv | Multi-Agent Framework. using the software collaboration as an example | DeepWisdom etc |  |
| Data Interpreter: An LLM Agent For Data Science | 2024 | Arxiv | address data science problem | DeepWisdom etc | |
| Agentless: Demystifying LLM-based Software Engineering Agents | 2024 | Arxiv | agentless method for swe-bench | UIUC |  |
| DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation | 2023 | ICML | code generation benchmark with a thousand data science questions | The University of Hong Kong etc |  |
| AutoCodeRover: Autonomous Program Improvement | 2024 | [Arxiv] | NUS |  |
|
| Name | Type | Target | Contributor | Code / Product |
|---|---|---|---|---|
| OpenDevin: Code Less, Make More | write code, fix bugs, and ship features. | SWE | OpenDevin Community |  |
| Devon: An open-source pair programmer | Codebase exploration Config writing Test writing Bug fixing Architecture exploration | SWE | entropy-research |  |
| gpt-engineer | write and execute software code | SWE | gpt-engineer-org |  |
| Aider is AI pair programming in your terminal | start a new project or work with an existing git repo. | SWE | paul-gauthier |  |
| Cover-Agent | automate and enhance the generation of tests (currently mostly unit tests) | QA Engineer | CodiumAI |  |
| PR-Agent | Automated Pull Request Analysis, Feedback, Suggestions | SWE | CodiumAI |  |
| GPT PILOT | VS Code extension that aims to provide the first real AI developer companion | SWE | Pythagora-io |  |
| Claude Engineer | assist with a wide range of software development tasks | SWE | Doriandarko |  |
| cognition AI | An applied AI lab building end-to-end software agents. | SWE | cognition.ai | Product |
| Tabby | self-hosted AI coding assistant, Github Copilot alternative | SWE | tabbyml |  |
| Sweep AI | issue PR, unit test | SWE | Sweep AI | Product |
| Continue AI | GitHub Copilot alternative, VS Code and JetBrains extension | SWE | Continue |  |
| Hex Magic AI | text to sql/python data analysis code, copilot to understand and fix code issues | MLE / DS | Hex | Product |
| datagpt | chatbot to SQL, automatic data analysis | DA | datagpt | Product |
| pandas-ai | Chat with your database (SQL, CSV, pandas, polars, mongodb, noSQL, etc). | DA | pandas AI |  |
| bito | generate, explain and review code | SWE | bito | [Product] |
| Name | Paper | Year | Blog | Institution | GitHub |
|---|---|---|---|---|---|
| Code Llama: Open Foundation Models for Code | arxiv | 2023 | link | Meta |  |
| CodeQwen1.5-7B | arxiv | 2024 | link | Alibaba |  |
| StarCoder2-Instruct: Fully Transparent and Permissive Self-Alignment for Code Generation | N.A. | 2024 | link | HuggingFace etc. |  |
| DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence | arxiv | 2024 | link | DeepSeek |  |
| Codestral | N.A. | 2024 | link | Mistral | [Hugging Face] |