⌘K
Change language Switch ThemeSign In
Narrow Mode
Stripe 工程师部署了自主智能体 Minions，每周生成数千个拉取请求
I InfoQ 中文 @InfoQ 中文
One Sentence Summary
Stripe has deployed an autonomous coding agent named Minions, which integrates LLMs with internal tools and blueprint workflows to automatically generate over 1,300 human-reviewed pull requests per week.
Summary
This article introduces Minions, an autonomous coding agent developed by Stripe. Unlike interactive assistants like Copilot, this system can complete software development tasks end-to-end based on Slack instructions, bug reports, or feature requests. The core of Minions utilizes a 'Blueprints' mechanism, combining deterministic code logic with flexible agent loops, and is deeply integrated with internal CI/CD pipelines to ensure code quality. Currently, Minions generates over 1,300 pull requests per week, excelling in well-defined tasks such as configuration adjustments, dependency upgrades, and small-scale refactoring, significantly boosting developer productivity.
Main Points
* 1. Minions marks the transition from interactive assistance to autonomous end-to-end task generation.Unlike traditional AI coding assistants, Minions can independently handle the entire process from task definition to PR submission, greatly reducing manual intervention by developers on routine and repetitive tasks. * 2. The 'Blueprints' mechanism balances process determinism with agent flexibility.By defining workflows through code, specifying how to break down tasks and invoke specific skills, it ensures that the agent's behavior remains reliable and controllable within a complex financial system supporting trillions of dollars in transactions. * 3. Deep integration with internal infrastructure is key to the successful implementation of AI agents.Minions is tightly coupled with Stripe's LLM infrastructure, version control, and CI/CD pipelines, enabling it to directly produce production-grade code that meets engineering standards.
Metadata
AI Score
87
Website mp.weixin.qq.com
Published At Today
Length 1395 words (about 6 min)
Sign in to use highlight and note-taking features for a better reading experience. Sign in now
Leela Kumili 2026-03-29 10:16 浙江
Stripe工程师介绍了Minions，这是一个自主编程的智能体，每周可生成超过1300个拉取请求。任务可以来自Slack、缺陷报告或功能需求。Minions利用大语言模型、蓝图和CI/CD流水线，生成可直接上线的代码变更，同时保证可靠性并保留了人工审核环节。
作者 | Leela Kumili
译者 | 张卫滨
Stripe 的工程师开发了 Minions，这是一个自主编程的智能体，能够根据单条指令端到端地完成软件开发任务。该系统实现了大语言模型（LLM）与内部开发者工具的集成，以极少的人工干预就能生成可直接用于生产环境的拉取请求（pull request）。Stripe 工程师表示，Minions 目前每周可生成超过 1300 个拉取请求，高于早期试验中的 1000 个。所有变更均会经过人工审核，但不包含任何人写的代码。
由 Minions 管理的代码支撑着 Stripe 每年超过 1 万亿美元的支付额，并运行在与金融机构、监管框架和合规性义务相关的复杂依赖关系中。在如此大规模的系统中部署自主智能体，可靠性与正确性仍是核心要求。
Stripe 的工程经理 Cameron Bernhardt 在 LinkedIn 的 帖子 中提到：
> Minions 已经从概念发展到每周生成上千个拉取请求。所有代码都会经过人工审核，但智能体正越来越多地实现端到端地变更生成。
Minions 与 GitHub Copilot 这类交互式编程助手或 Cursor 这类 AI 代码编辑器不同，它执行 一次性)、端到端的任务。任务可以来自多个源，包括 Slack 对话、缺陷报告或功能需求。任务描述会发送给 Minion，然后 Minion 使用蓝图（一组确定性代码与灵活智能体循环的组合）来编排工作，生成所需的代码、测试和文档。工作流最终会提交一个拉取请求供人工审核，这样工程师从任务定义到可上线的代码只需极少的手动操作。
调用 Minion 的 Slack 消息示例（来源：Stripe 的博客文章）
该系统由 Block 开发的 Goose 的内部分支演进而来，Goose 是最早得到广泛应用的编程智能体之一。针对 Stripe 的 LLM 基础设施，Goose 进行了定制，并为满足 Minions 的特定需求做了优化，而 Cursor、Claude Code 等交互式工具继续支持人工监督的工作流。
Minions 通过蓝图（blueprints）进行编排，蓝图是用代码定义的工作流，规定如何将任务拆分为子任务，并根据任务类型由确定性的例程（routine）或智能体来进行处理。Stripe 工程师将蓝图描述为与代码交织在一起的智能体技能集合，在保证效率的同时保留了适应性。
Minion 示例蓝图（来源：Stripe 的博客文章）
系统通过 CI/CD 流水线、自动化测试和静态分析来强化可靠性，确保生成的变更在人工审核前符合工程标准。工程师指出，Minions 在定义明确的任务上表现最佳，例如配置调整、依赖升级和小型重构。
该系统反映了智能体驱动软件开发的广泛趋势：基于 LLM 的智能体与开发环境、版本控制和 CI/CD 流水线深度集成，以极少的监督生成生产级的代码。Stripe 的经验表明，自主编程智能体可以显著提升开发者的生产力，同时保持严格的质量控制。
查看英文原文：
Stripe Engineers Deploy Minions, Autonomous Agents Producing Thousands of Pull Requests Weekly（https://www.infoq.com/news/2026/03/stripe-autonomous-coding-agents/）
声明：本文为 InfoQ 翻译，未经许可禁止转载。 点击底部**阅读原文访问 InfoQ 官网，获取更多精彩内容！****
今日好文推荐 谷歌迎来“DeepSeek时刻”！TurboQuant引爆AI圈、全球开发者疯狂复现：6倍无损压缩，内存股集体暴跌 1.9万行 Claude Code“AI垃圾”杀入 Node.js：全球顶流开源项目，快守不住了 遭谷歌封禁，OpenClaw 创始人首次在 OpenAI 受访吐槽：Gemini 自信“100%能跑”，结果一试就崩 “Token 不用完会焦虑”！Karpathy最新访谈自曝患上 “AI 精神病”：软件世界正在被 Agent 接管
I InfoQ 中文 @InfoQ 中文
One Sentence Summary
Stripe has deployed an autonomous coding agent named Minions, which integrates LLMs with internal tools and blueprint workflows to automatically generate over 1,300 human-reviewed pull requests per week.
Summary
This article introduces Minions, an autonomous coding agent developed by Stripe. Unlike interactive assistants like Copilot, this system can complete software development tasks end-to-end based on Slack instructions, bug reports, or feature requests. The core of Minions utilizes a 'Blueprints' mechanism, combining deterministic code logic with flexible agent loops, and is deeply integrated with internal CI/CD pipelines to ensure code quality. Currently, Minions generates over 1,300 pull requests per week, excelling in well-defined tasks such as configuration adjustments, dependency upgrades, and small-scale refactoring, significantly boosting developer productivity.
Main Points
* 1. Minions marks the transition from interactive assistance to autonomous end-to-end task generation.
Unlike traditional AI coding assistants, Minions can independently handle the entire process from task definition to PR submission, greatly reducing manual intervention by developers on routine and repetitive tasks.
* 2. The 'Blueprints' mechanism balances process determinism with agent flexibility.
By defining workflows through code, specifying how to break down tasks and invoke specific skills, it ensures that the agent's behavior remains reliable and controllable within a complex financial system supporting trillions of dollars in transactions.
* 3. Deep integration with internal infrastructure is key to the successful implementation of AI agents.
Minions is tightly coupled with Stripe's LLM infrastructure, version control, and CI/CD pipelines, enabling it to directly produce production-grade code that meets engineering standards.
Key Quotes
* Minions has evolved from a concept to generating thousands of pull requests per week. All code is reviewed by humans, but the agent is increasingly achieving end-to-end change generation. * Minions is different from interactive coding assistants like GitHub Copilot or AI code editors like Cursor; it executes one-off, end-to-end tasks. * Stripe's experience shows that autonomous coding agents can significantly boost developer productivity while maintaining strict quality control.
AI Score
87
Website mp.weixin.qq.com
Published At Today
Length 1395 words (about 6 min)
Tags
Stripe
Minions
AI Agent
Autonomous Coding
LLM
Related Articles
* “AI on the Front Lines: How Developers are Reshaping the Software Development Process” | Roundtable Discussion * 128. Manus Co-founder's Final Interview Before Sale: The Fantastical Drift to 2025... * From Context to Long-Term Memory: Architectural Design and Practice of LLM Memory Engineering\" architecture.") * 1,500 PRs, 0 Human Coders: Building a Million-Line Internal Product Driven by Codex * OpenAI Frontline Development Observations: Those Who Can Manage 10-20 Agents Simultaneously and Run Hour-Long Tasks Are Leaving Other Engineers Far Behind * How to Design an AI Agent System * Deconstructing Clawdbot: Local Architecture, Memory Management, Agent Orchestration, and Context Assembly Principles * GPT-5.4 Released: OpenAI's First Unified Model, Truly Native * Practices and Reflections on Vibe Coding in Code Generation and Collaboration * What You Don't Know About Claude Code: Architecture, Governance, and Engineering Practices - Tw93 HomeArticlesPodcastsVideosTweets