Title: 智能体 AI 模式强化软件工程规范性 | BestBlogs.dev
URL Source: https://www.bestblogs.dev/article/cfedb41a
Published Time: 2026-04-05 02:15:00
Markdown Content: Skip to main content Toggle navigation menu Toggle navigation menuArticlesPodcastsVideosTweetsSourcesNewsletters
⌘K
Change language Switch ThemeSign In
Narrow Mode
智能体 AI 模式强化软件工程规范性
I InfoQ 中文 @InfoQ 中文
One Sentence Summary
This article explores how agentic AI patterns strengthen and evolve software engineering disciplines in AI-assisted environments through mechanisms such as spec-driven development, automated validation, and shift-right feedback.
Summary
This article synthesizes insights from industry experts like Paul Duvall, Paul Stack, and Gergely Orosz, highlighting that with the surge in AI-generated code, traditional engineering disciplines (such as CI, automated testing, and XP practices) have become more, not less, important. Key points include: adapting mature engineering practices using agentic AI patterns; shifting from inefficient manual line-by-line reviews to relying on agentic guardrails and automated validation; promoting spec-driven development to address AI intent ambiguity; and using 'shift-right' practices to feed real-time production signals back into the development lifecycle. Furthermore, experts predict that R&D teams will transition toward smaller, highly automated 'two-pizza teams.'
Main Points
* 1. In the context of AI-generated code, the importance of traditional engineering practices has increased rather than decreased.As the speed of code changes accelerates, trunk-based development, frequent commits, and automated testing have become the cornerstones of quality assurance, preventing massive amounts of AI-generated content from causing system instability. * 2. The developer role is shifting from 'code writer' to 'spec definer' and 'automated supervisor'.Manual line-by-line review is no longer practical; developers should rely on structured specifications to drive AI, utilizing agents for self-review, remixing, and automated validation. * 3. Spec-driven development is key to resolving AI intent ambiguity.By using predefined roles, context, constraints, and acceptance criteria, we ensure that agents generate and validate output based on clear specifications, avoiding random results caused by ambiguous inputs. * 4. Software development processes are evolving toward 'shift-right' feedback and smaller team models.Leveraging telemetry data from production environments shortens feedback loops, while AI reduces coordination costs, driving R&D teams toward more efficient, focused 'two-pizza teams'.
Metadata
AI Score
87
Website mp.weixin.qq.com
Published At Today
Length 2110 words (about 9 min)
Sign in to use highlight and note-taking features for a better reading experience. Sign in now
InfoQ 2026-04-05 10:15 北京
Paul Duvall 介绍了其面向 AI 辅助开发的工程模式库与高质量交付实践；Paul Stack 和 Gergely Orosz 则指出，行业正转向重构整合与规范驱动开发。
作者 | Rafiq Gemmail
译者 | 明知山
在最新一期的 AI DevOps 播客 中，Paul Duvall 探讨了随着现代模型能力的提升，智能体 AI 模式如何强化核心工程规范。他还分享了自己的 智能体 AI 工程模式 代码库，用于记录和演进 AI 辅助软件开发实践。
Duvall 是“Continuous Integration: Improving Software Quality and Reducing Risk”一书的作者，他将这一系列模式定位为：探索如何通过客户在实际工作中应用智能体 AI 的实践经验来适配已成熟的工程实践。他强调应让 AI 生成的内容基于共享模式，并指出：“当 AI 可以生成代码，工程实践比以往任何时候都更为重要。”
面对 AI 生成的海量代码，Duvall 强调了主干开发、尽早且频繁提交以及自动化测试的重要性，并解释说，随着变更速度加快，这些实践对保障质量至关重要。
Duvall 还描述了开发者与代码交互方式的转变。他发现，在处理 AI 生成的内容时，自己“不再逐行审查代码”，因为海量变更让这种方式越来越不切实际。对此，Duvall 强调应该更多地依靠自动化验证与智能体防护机制，包括让智能体具备审查和优化自身输出的编码能力。
Duvall 还探讨了规范驱动开发等方法是怎样让现有软件工程实践不断演进的。他的代码库中包含适用于 AWS IAM 策略生成场景的智能体可读规范示例，通过预先定义预期行为、约束条件与验收标准，让智能体能够依据清晰的规范生成并校验输出。在谈及经典的测试优先模式如何被用来指导 AI 辅助工作流时，他表示：
> 我其实……是在复用我们在敏捷开发和极限编程（XP）中的做法……说白了就是先红、再绿、再重构……我就是按这个流程在做。
Duvall 还指出了智能体生命周期初期阶段存在的挑战，尤其是在意图定义环节。他发现，尽管 AI 工具能够快速生成代码，但模糊、不明确的输入往往会导致结果不一致且难以预测。这也使得人们更加重视通过更清晰的规范来驱动智能体，包括使用包含角色、上下文与约束的结构化提示词来描述意图，以及采用规范驱动开发和基于行为定义的验收测试。他表示：“如果没有完整地描述清楚意图，就只会得到随机的结果。”
在近期的 DevSecOps Talks 播客中，System Initiative 产品总监 Paul Stack 也谈到了对更清晰规范的重视。他主导开发的 SWAMP 是一个用于自动化与验证基础设施的开源智能体平台。Stack 介绍了自己如何围绕智能体重构开发流程，甚至放弃拉取请求，转而采用基于 GitHub Issue 的工作流，并融入规范驱动开发模式。他表示：
> 我们不接受拉取请求……如果你有设计方案，请提交一个 Issue，我们会进行交互式的合作，一起完善、共同设计。
在 Scott Hanselman 的播客中，“The Pragmatic Engineer” 新闻组作者 Gergely Orosz介绍 了一个开源项目：该项目不合并拉取请求，而是采用 “remixing” 模式，由智能体按照项目标准重构提交的 PR。Hanselman 认为，尽管架构与设计“品味”对系统来说至关重要，但“拥有无限耐心的初级工程师”这种心态很适合用来处理繁琐重复性的工作，这与完全自动化的“Ralph 循环”自主智能体形成对比——这类智能体通过子智能体迭代优化方案，直至满足需求。
Stack 还强调必须提供准确的架构模式与实践规范，让智能体能够“以与代码库一致的方式生成代码”，以及预先定义架构、约束条件与测试预期。与 InfoQ 报道的 Boris Cherny 智能体工作流 类似，Stack 表示他也会使用 Claude 的“计划模式”在执行前对意图进行审查，以此来避免出现“AI 恐怖故事”式的问题。
Duvall 还指出了右移实践（Shifting-Right）的重要性，主张将这类反馈循环延伸至生产环境。他阐述了如何利用可观测性、遥测数据乃至生产环境中的测试来缩短反馈周期，解析实时信号并回送至开发生命周期。展望未来，他认为 AI 可能催生出规模更小、目标更聚焦的团队，并提到随着协调成本降低、自动化程度提升，研发团队将向“单披萨团队”（One-Pizza Team）模式转型。
Duvall 认为，与以往的工程变革一样，质量保障将越来越依赖自动化，而非人工审查。他表示：
> 你正在搭建相应机制……让代码得到审查……只不过未必每次都由你亲自审查。
Duvall 与 Stack 均强调，AI 辅助开发需要结合左移实践与右移反馈，将行为定义与生产状态纳入验证流程。Duvall 还指出，AI 能够更全面地分析生产遥测数据，从而识别出模式、更早地发现问题。
Duvall 的代码库在持续更新中，其中定义了涵盖开发、安全与运维场景的结构化模式及成熟度等级。这些模式包括 规范驱动开发、编码规则 和 架构约束、原子分解 与 并行智能体，以及具备 自动化可追溯性 的 可观测开发。
Orosz 意识到开发正在向以代码为中心之外的方向转变，他认为工程的定位与实践将会提升到新层次，超越代码本身。他表示：
我认为，除了编写代码之外，还有一些特质会让我们与众不同，我们理应去培养这种能力。
查看英文原文： https://www.infoq.com/news/2026/03/agentic-engineering-patterns/
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 Anthropic 掐断“龙虾”补贴，OpenClaw 创始人哭求仅换7天续命，网友吵疯：作秀还是义务 OpenAI 正在做一个“替你用电脑”的 Super App，新模型 Spud 几周内登场 谷歌重磅开源Gemma 4！手机离线跑 Agent、还降内存，Qwen 被拉进正面对决 一个周末 + 1100 美元，干完 5 人 6 个月的活：Cloudflare 用 AI“复刻”Next.js，已跑进生产环境
I InfoQ 中文 @InfoQ 中文
One Sentence Summary
This article explores how agentic AI patterns strengthen and evolve software engineering disciplines in AI-assisted environments through mechanisms such as spec-driven development, automated validation, and shift-right feedback.
Summary
This article synthesizes insights from industry experts like Paul Duvall, Paul Stack, and Gergely Orosz, highlighting that with the surge in AI-generated code, traditional engineering disciplines (such as CI, automated testing, and XP practices) have become more, not less, important. Key points include: adapting mature engineering practices using agentic AI patterns; shifting from inefficient manual line-by-line reviews to relying on agentic guardrails and automated validation; promoting spec-driven development to address AI intent ambiguity; and using 'shift-right' practices to feed real-time production signals back into the development lifecycle. Furthermore, experts predict that R&D teams will transition toward smaller, highly automated 'two-pizza teams.'
Main Points
* 1. In the context of AI-generated code, the importance of traditional engineering practices has increased rather than decreased.
As the speed of code changes accelerates, trunk-based development, frequent commits, and automated testing have become the cornerstones of quality assurance, preventing massive amounts of AI-generated content from causing system instability.
* 2. The developer role is shifting from 'code writer' to 'spec definer' and 'automated supervisor'.
Manual line-by-line review is no longer practical; developers should rely on structured specifications to drive AI, utilizing agents for self-review, remixing, and automated validation.
* 3. Spec-driven development is key to resolving AI intent ambiguity.
By using predefined roles, context, constraints, and acceptance criteria, we ensure that agents generate and validate output based on clear specifications, avoiding random results caused by ambiguous inputs.
* 4. Software development processes are evolving toward 'shift-right' feedback and smaller team models.
Leveraging telemetry data from production environments shortens feedback loops, while AI reduces coordination costs, driving R&D teams toward more efficient, focused 'two-pizza teams'.
Key Quotes
* When AI can generate code, engineering practices are more important than ever. * I'm actually... reusing what we did in Agile development and Extreme Programming (XP)... simply put, it's red, green, refactor. * If you don't describe your intent clearly and completely, you'll just get random results. * You are building mechanisms... to get code reviewed... it just doesn't necessarily have to be reviewed by you personally every time. * Beyond writing code, there are other traits that set us apart, and we should cultivate those capabilities.
AI Score
87
Website mp.weixin.qq.com
Published At Today
Length 2110 words (about 9 min)
Tags
Agentic AI
Software Engineering
Spec-driven Development
Automated Testing
DevOps
Related Articles
* Cognitive Reconstruction: After Three Months with Speckit, I Abandoned It - Escaping the Dilemma of Powerful Tools That Are Hard to Use Well * 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 * Practices and Reflections on Vibe Coding in Code Generation and Collaboration * From 7.9% to 54% Adoption: The Three-Stage Evolution of Kuaishou's Intelligent Code Review * “AI on the Front Lines: How Developers are Reshaping the Software Development Process” | Roundtable Discussion * Deconstructing Agentic Coding from First Principles: From Theory to Practice * From Context to Long-Term Memory: Architectural Design and Practice of LLM Memory Engineering\" architecture.") * From Clawdbot to the 2026 AI Coding Explosion | A Conversation with PingCAP CTO Dongxu * What You Don't Know About Claude Code: Architecture, Governance, and Engineering Practices - Tw93 HomeArticlesPodcastsVideosTweets