Tag
作者认为AI驱动的软件开发中，评测系统（evals）正在取代传统自动化测试，成为提升开发效率和系统质量的关键，并提到Claude Code团队和Google的harness2等大厂都在加大投入。
观察到许多独立开发产品的开发者已停止直接使用 git，而是让 Claude 或 Codex 代办所有版本控制工作，传统的开发"肌肉记忆"正在迅速消退。
The paper introduces TokenEconomizer, an experience-driven external controller that reduces token costs of coding agents by injecting micro-skills, pruning irrelevant tools, and interrupting failure loops, achieving 18.4–30.9% savings across Claude Code, Codex, and GitHub Copilot CLI with comparable task success.
该论文分析了仓库级编码基准中的过度规范与规范不足问题，指出将真值信息泄露到任务描述会形成"剧透"，并提出生成简洁提示的建议以提升基准的可靠性与持久性。
Kent C. Dodds notes his codebase has reached 1.2 million lines of code while quoting MEGA's take that humans—not tools—become the bottleneck in agent-driven development, pointing to async, remote agents as the architectural fix.
GitHub blog 分享在 AI 改变开发工作方式的背景下，开发者应重点培养的三项技能：学会指挥 AI 代理而非仅使用它、不盲目信任 AI 的首个答案、以及利用省下的时间解决更大的问题。
The author argues that fears about AI replacing software engineers and mathematicians are misplaced, claiming those professions are being automated first because they are the most modern and transparent, while obsolete administrative and manual jobs will follow once processes are updated.
patio11 附和 Sahil Lavingia 关于「政府软件黄金时代」的观点，但提出限定条件：那些只是因为缺乏工程投入而糟糕的政府应用将迎来黄金时代，而那些因政策决策困难而破损的应用，仅仅换个新界面是无济于事的。
A tweet humorously lists essential engineering concepts vibe coders lack, citing a case where a South African coffee chain CEO's vibe-coded dashboard leaked sensitive data in its login source code.
The "Devin Meetup Osaka #2," sharing real-world use cases and practical know-how for the autonomous AI engineer "Devin," will be held on October 23 at MonotaRO's head office. Through sessions, lightning talks, and a networking event, participants will share tips on crafting effective prompts, code review and quality assurance, and best practices for introducing Devin to teams.
一篇关于「脑三明治」工作流的博客文章，提出工程师在使用 AI 编码时应先用脑理解问题、再放手让 AI 执行、最后再用脑审查成果，以避免技能退化并保持代码可审查性。
Claude Code 的开发流程表明，为 AI 代理提供自动化验证机制（如 E2E 测试、属性测试和多代理对抗审查）可让 Claude 自行编写 90% 以上的代码，同时在 PR 合并量翻倍的情况下将回滚率降低 40% 以上。
A tweet criticizing the overhyped notion that LLMs have 'solved' writing extremely complex programs, using verified compilers like CompCert as a counterexample of a task that still demands top-tier human expertise.
Microsoft and Carnegie Mellon University launched the PRAISE program to train early-career Microsoft software engineers in AI-augmented software engineering through online CMU courses and on-the-job mentorship. The initiative aims to help junior engineers adapt to AI tools reshaping the profession.
一条观点推文，作者认为工程师若不全力投入 AI 就不算真正的工程师，呼吁放下自我、让 AI 接管控制权。
The author introduces 'Anchor Engineering,' a software development method that encodes human design judgement into 'anchors'—guiding constraints stored in an 'Anchor Book' alongside the code—so AI coding agents can build systems faster without losing shared understanding, quality, or team coordination.
Gergely Orosz reflects on an unprecedented moment where the entire tech industry—from startups to big tech, nonprofits, VCs, government, and academia—is simultaneously rethinking developer tools and processes, largely driven by AI.
开发者在2026年就业市场中常犯的错误是把完成项目等同于工程能力证明；AI降低了构建软件的门槛，但没有降低理解软件和做出正确决策的标准。作者强调面试中能解释架构决策和权衡取舍比展示AI生成的项目功能更重要。
一名工程师分享个人感悟：意识到自己并不比 AI 更聪明，未来应专注于成为 AI 的优秀编排者，为其提供正确的工具、数据与权限，同时承认自己比 AI 更懂的领域正在快速缩小。
A retrospective 1982 piece speculating on the future of programming, reflecting the era's predictions about languages, tools, and software development.