AI编程助手正在帮助开发者更高效地编写代码。但为什么大多数企业实际上并没有交付更多软件？这 是GitLab CEO Bill Staples在一年多前接任该职位后，不断从客户那里听到的问题。 "他们会说：我们已经投资了，正在使用这些编程工具。我们的工程师很喜欢它们，但我们没有看到创 新速度的加快，"Staples在与客户交谈时听到这样的反馈，"我们没有更快地交付更多软件。" 在本期The New Stack Makers节目中，我们与Staples坐下来讨论了为什么编程从来不是真正的瓶颈， GitLab新发布的Duo智能体平台如何旨在自动化完整的软件开发生命周期（SDLC），以及为什么上下 文而不仅仅是代码生成，是让智能体AI在企业中发挥作用的关键。 编程不是瓶颈 拥抱创新生态系统 在这个快速变化的环境中，许多企业必须问自己应该投注于哪些工具。 Staples说他在担任CEO的前100天里与60多位客户进行了交谈。他惊讶地发现，即使是像金融服务和公 共部门这样高度监管的企业也在全力投入AI。但即使他们全力投入AI编程工具，软件交付速度并没有 明显加快。 正如Staples所指出的，开发者每天只花费10%到20 ...

然而在亚马逊内部，员工若未获得正式批准，则不得将 Anthropic 的 Claude Code 用于编写代码、开发 产品等实际生产环境。并且这一矛盾在去年秋天开始变得更加明显，亚马逊当时发布了内部指引，建议 员工优先使用自研 AI 编程助手 Kiro。 据悉，Kiro 虽然基于 Claude 研发，但能够配合亚马逊云服务（IT之家注：AWS）等自有工具。这一政 策在内部论坛里引来轩然大波，不少员工对此表示批评。 并且，一些负责销售第三方 AI 服务的员工完全不理解这项政策。他们平时的工作就是把 Claude Code 等产品卖给客户，一些员工质疑道："如果我们连正式工作中都不能用 Claude Code，那如何向客户推广 这些产品呢？或者说，客户应该如何信任一个我们自己都没有批准使用的工具？"。 IT之家 2 月 12 日消息，据《商业内幕》昨天报道，亚马逊已在内部优先推广自研 AI 编程工具 Kiro， 限制员工使用 Claude Code 等第三方产品。 据报道，亚马逊是 Anthropic 的大股东之一，也帮助这家初创 AI 公司将其 AI 模型、产品推向 C 端（客 户端）市场。 其余工程师也讨论了 ...

Company and Industry Summary Company: 智谱 (02513.HK) Key Points 1. **Model Iteration and Launch** 智谱 recently released and open-sourced GLM-4.7-Flash, which replaces GLM-4.5-Flash. The model is positioned for "lightweight deployment + strong inference/coding" [1] 2. **PonyAlpha Testing and Feedback** OpenRouter launched the stealth model PonyAlpha on February 6, which supports 200K context. Feedback from the overseas community highlights its outstanding capabilities in "coding/agentic workflows," suggesting it may be the next generation of GLM [1] 3. **Developer Market Potential** An official report indicates that the number of software developers in China has surpassed 9.4 million. The Ministry of Industry and Information Technology disclosed that software business revenue is projected to reach 137.276 billion yuan (approximately 13.7 trillion yuan) in 2024. Even a single-digit penetration of AI programming assistants/agents in the toolchain corresponds to a potential market space in the hundreds of billions [2] 4. **Client Base and Revenue Drivers** According to the company's prospectus, it serves over 8,000 institutional clients. Nine out of the top ten internet companies in China are using GLM models, and the quality of government and enterprise clients is high, indicating strong customer retention and willingness to pay [2] 5. **Cloud Business Revenue and Profitability** The company's cloud business revenue and gross profit are expected to continue to show elasticity as model capabilities improve [3]

除了文本输入，用户还可以上传界面草图等多模态内容作为开发参考。 01 应用的一个重要功能是最近很流行的"技能"(Skills）系统。这是一套用户可自定义的扩展机制，用于调整Codex的输出 方式。 OpenAI在2月3日正式发布了其编程助手Codex的macOS桌面应用。这款应用目前向所有ChatGPT用户开放试用，包括 免费用户和Go餐用户。对于付费订阅用户，OpenAI还临时提供了双倍的使用额度。 Codex应用的界面设计相当简洁，核心是一个类似ChatGPT的对话框。用户输入需求后，应用会逐条显示生成的代码 片段。对话框旁边有一个侧边栏，提供项目文件链接等快捷入口。 整个交互逻辑围绕对话展开，开发者可以通过语言描述需求，让agent完成具体的编程工作。 根据siliconangle的实测，Codex桌面应用版的底层模型仍然是2025年12月发布的GPT-5.2-Codex模型。该模型可以处理 最多40万个token输入，相当于约10万行代码的规模，并且支持超过50种编程语言。 比如开发者想要直观了解进度，那就给Codex添加新技能，在每次建议代码修改时自动生成配套的可视化说明。 创建技能的过程就是组织一 ...

Core Insights - Feisuan Technology has officially launched the JavaAI Professional Edition, recognized as the only AI programming assistant certified by the China Academy of Information and Communications Technology for generating complete engineering code [2] - Since its introduction in early 2025, Feisuan JavaAI has assisted Java developers in generating over 1 million complete projects [2] Performance Improvements - The JavaAI Professional Edition has achieved systematic breakthroughs in model capability, code quality, and usage efficiency [2] - Code adoption rate has increased from 70% to 90% [2] - Generation speed has improved by 30% [2] - Rework and debugging time has been reduced by 20% [2] Tool Upgrades - The launch includes upgrades to ten AI tools that cover the full range of needs for Java developers, including code repair, test generation, and security protection [2] - These tools are designed to meet the daily work scenarios of Java developers, ensuring production readiness [2]

Core Viewpoint - The Linux kernel community is debating whether to establish a specific submission guideline for "tool-generated code," particularly concerning AI programming assistants and LLM-generated patches, amid concerns about the influx of low-quality "AI-generated patches" known as AI Slop [1][3]. Group 1: Linus Torvalds' Position - Linus Torvalds emphasizes that documentation should focus on the tools themselves rather than targeting AI directly, as AI-assisted submissions will persist regardless of documentation [1][3]. - He criticizes the notion that AI-generated code can be effectively labeled or regulated through documentation, stating that those submitting low-quality AI code are unlikely to mark it as such [3][5]. - Torvalds dismisses the idea that documentation can solve the issue of AI-generated garbage code, labeling such discussions as naive and ineffective [3][5]. Group 2: Community Perspectives - There are two extreme viewpoints within the community: one side believes AI will destroy software engineering, while the other sees it as a revolutionary force for automation [5]. - Torvalds maintains a neutral stance, insisting that the only appropriate characterization of AI in documentation is as a tool, avoiding any divisive rhetoric [5][6]. - The debate reflects a broader anxiety within the Linux community about the future of development practices and the role of AI, rather than merely a technical specification issue [5][6]. Group 3: Governance and Code Quality - Torvalds argues that rules can only constrain those who are already compliant, and those who wish to submit low-quality patches will ignore any guidelines, regardless of their length [4][5]. - He asserts that the real focus should be on code review mechanisms, the judgment of maintainers, and the community culture, which cannot be automated or regulated through documentation [6].

Core Insights - The article discusses the evolution of AI programming assistants, highlighting the shift from simple code completion tools to more advanced models capable of understanding complex tasks and contexts. This evolution is represented by two main routes: IDE enhancement and Agentic coding [1][2]. Group 1: AI Programming Assistant Evolution - AI programming assistants have significantly changed development workflows, with even skeptics like Linus Torvalds acknowledging their utility [1]. - The article identifies two main routes for AI programming assistants by 2025: IDE enhancement (e.g., GitHub Copilot) and Agentic coding (e.g., Claude Code) [2]. Group 2: Doubao-Seed-Code Introduction - Doubao-Seed-Code, developed by Volcano Engine, aims to address the limitations of existing models by providing a robust programming model designed for complex tasks [2][4]. - The model has shown exceptional performance in various authoritative benchmarks, even surpassing Claude 4.5 Sonnet in some evaluations [6][8]. Group 3: Key Features of Doubao-Seed-Code - Doubao-Seed-Code boasts a native 256K long context capability, allowing it to handle complex projects that span multiple files and dependencies [10][11]. - The model is the first in China to support visual understanding, enabling it to generate code based on UI designs and perform visual comparisons for style and bug fixes [11]. Group 4: Performance Evaluation - The article outlines a series of practical tests to evaluate Doubao-Seed-Code's capabilities, including task planning, long context handling, and debugging abilities [18][22]. - In a test involving the refactoring of a poorly structured Python script, Doubao-Seed-Code completed the task in under three minutes, demonstrating its debugging capabilities [23][24]. Group 5: Advanced Task Execution - Doubao-Seed-Code successfully executed a complex task of converting a C++ game to Python, showcasing its long context and task planning abilities. The entire process took approximately 40 minutes [26][30]. - The model autonomously planned and executed the project, demonstrating its capability to handle significant programming challenges [31]. Group 6: Cost and Accessibility - Doubao-Seed-Code aims to address pricing and usage limitations faced by developers, offering a subscription service with competitive pricing [48][50]. - The "Coding Plan" subscription service provides significant discounts and aims to lower costs by 62.7%, making it accessible to a broader range of developers [49][50]. Group 7: Conclusion - Doubao-Seed-Code is positioned as a powerful alternative in the Agentic coding space, capable of handling complex tasks autonomously and efficiently [52][53]. - The model not only addresses performance issues but also offers a cost-effective solution for developers, paving the way for widespread adoption of Agentic coding [53][54].

Core Insights - Anthropic has launched the web version of its AI programming assistant, Claude Code, making coding more accessible by eliminating the need for command-line tools and complex commands [2][5] - The web version is currently in testing and available only to Pro and Max subscribers, aimed at gathering user feedback for further improvements [6] - Claude Code has seen a tenfold increase in users since its broader release in May, generating over $500 million annually for Anthropic [27] Group 1: Claude Code Features - The web version allows users to initiate programming tasks directly through a browser, connecting to GitHub repositories and describing task requirements for Claude to handle automatically [12][13] - Claude Code can process multiple tasks in parallel, providing real-time progress tracking and the ability to guide the AI during task execution [14] - The cloud-based execution ensures tasks run in isolated environments, enhancing security by limiting access to authorized repositories and allowing custom network configurations [16] Group 2: Claude for Life Sciences - Anthropic has introduced Claude for Life Sciences, utilizing the Claude Sonnet 4.5 model, which outperforms human averages in experimental protocol understanding [20] - This version includes specialized connectors for direct integration with experimental platforms, databases, and literature, enabling Claude to function as a research assistant [21][22] - The new Agent Skills feature allows Claude to execute specific tasks autonomously, enhancing its capabilities in scientific research [23] Group 3: Market Impact and Growth - Anthropic's valuation has reached $183 billion, reflecting its significant market presence and growth potential [28] - The introduction of Claude Code and its rapid user growth indicate a strong demand for AI-driven programming solutions [27]

Core Insights - The article highlights the rapid growth and funding success of Blacksmith, a startup focused on accelerating code deployment processes in the AI-driven software industry [1][2]. Funding and Performance - Blacksmith secured a $10 million Series A funding round led by Google Ventures, completing the deal in just 14 days, following a $3.5 million seed round in May [2]. - The company has seen significant growth in annual recurring revenue (ARR), increasing from $1 million to $3.5 million after expanding its team from 4 to 8 members and acquiring over 700 customers [3]. Market Position and Services - Blacksmith offers continuous integration and continuous delivery (CI/CD) services that complement GitHub Actions, addressing the high costs and unpredictability associated with software deployment [2][3]. - The startup operates on high-performance gaming-grade CPUs, achieving up to 2x processing speed and reducing computing costs by up to 75% compared to traditional cloud service providers [4][6]. Business Model and Strategy - By utilizing a bare-metal architecture, Blacksmith claims to have better economic control than large-scale cloud providers, allowing for improved profit margins as they scale their customer base [6]. - The target customers are enterprises with engineering teams of over 500, and notable clients include Ashby, Chroma, and Supabase [6][7]. Team and Background - The founding team of Blacksmith has experience from Faire and Cockroach Labs, where they recognized the challenges in the software release process [3]. - The company graduated from Y Combinator's winter 2024 batch and currently has a team of 11 members [7].

Core Insights - OpenAI has acquired the popular Xcode plugin Alex, with the entire team joining Codex, which may significantly impact Codex's presence in the Mac development ecosystem [1][15]. Group 1: Alex Plugin Overview - Alex is recognized as the best Copilot for Xcode, a powerful IDE for Apple platforms, and has a high market share among Mac developers [1]. - The plugin integrates deeply with Xcode through macOS accessibility interfaces, enabling features like code auto-completion, conversational suggestions, and automated development tasks [3]. - Within a week of its launch, Alex accumulated over 1,200 active users, including developers from notable companies like Notion and LinkedIn [3][15]. Group 2: Team Background - The Alex team consists of only three members, led by founder Daniel Edrisian, a young and accomplished iOS developer [4]. - Edrisian has a strong background in iOS development, having received multiple Apple WWDC student scholarships and interned at major tech companies [6][11]. Group 3: Strategic Implications for OpenAI - The acquisition of Alex allows OpenAI to expand its influence in the Apple ecosystem, an area where Codex has been relatively underrepresented [15]. - By integrating Alex's technology and expertise, OpenAI aims to enhance its support for Apple app development, leveraging Alex's proven market demand [15]. - This partnership presents a win-win scenario, providing Alex's team with access to OpenAI's extensive resources while enhancing Codex's capabilities [15].