小程序上线半年我赚了多少钱？
The author reflects on the journey of developing and launching "Fuzhu Accounting," a mini-program, over the past six months. Starting from version 1.0.0 and progressing to 1.6.1, the project was completed in the author's spare time, marking a significant personal achievement and their first step into independent development. Despite minimal promotion, the mini-program accumulated 872 users, primarily through natural traffic after a single WeChat official account article. However, user retention remains a challenge, with daily active users stabilizing at around 10-20, prompting the author to seek user feedback and continuously optimize the product.
Financially, the mini-program's revenue is currently negligible, totaling only 69.32 (presumably RMB) over three months from a single splash screen ad, as all features are free. The author acknowledges that a lack of revenue is unsustainable and plans to introduce VIP features once the product is more complete and robust. The article delves into the author's insights on independent development, emphasizing the difference between creating a demo and a full product, the importance of product and operations thinking over pure programmer mindset, and the transformative role of AI in accelerating development.
Crucially, the author highlights the importance of product positioning and marketing, especially in competitive "red ocean" markets like accounting apps. They advocate for differentiation through unique features, UI, or interaction, focusing on "small and beautiful" solutions for specific user needs rather than attempting a comprehensive approach. Without a marketing budget, self-promotion and building a personal brand become vital. The author concludes by embracing the unknown path of independent development as a journey filled with infinite possibilities.
是江迪呀 • 2026-01-10 • 48,055 views
微信小程序 产品 创业
Read original on juejin.cn →
Cursor 转 Codex 大半个月，聊聊我的真实感受
The author recounts their journey with AI in coding, starting in April 2025 when AI became commercially viable. They initially subscribed to Cursor, an AI-powered IDE, which their company later adopted. Cursor proved to be a revolutionary tool, deeply integrating AI features like Tab completion, Cmd+K, and Agent modes directly into the coding environment, significantly enhancing productivity and reducing manual coding efforts. Despite some initial technical quirks, Cursor's rapid iterations and the advancements in underlying AI models (like Claude) quickly elevated its capabilities, making it indispensable for complex business logic and even enabling the author to develop a complete mini-program with zero code.
深小乐 • 2026-05-10 • 28,963 views
人工智能
Read original on juejin.cn →
体验完阿里首款Design Agent，我开始替UI/前端焦虑了..
The author, a backend developer with no UI design expertise, recounts his long-standing struggle to create visually appealing and interactive frontend pages despite his technical skills. This challenge led him to consider learning UI design and tools like Figma. However, his recent experience with Alibaba's QoderWork AI Native Design Desk, an AI Design Agent, has completely changed his perspective, leading him to believe that the traditional roles of UI designers and frontend developers might be facing significant disruption.
Using QoderWork, the author, in just 20 minutes, generated a highly professional and interactive product website that he believes surpasses many existing company websites. He highlights the tool's intuitive features, such as direct canvas editing for real-time adjustments, box selection for targeted modifications, and exposed global design parameters for quick, live theme changes. QoderWork's ability to understand natural language prompts, ask clarifying questions, and output complete React + Vite project files (unlike tools that only generate HTML) makes it accessible even to non-developers, while also offering advanced capabilities for professionals.
The article concludes with the author's personal anxiety regarding the future of UI and frontend roles, suggesting that AI-driven design agents significantly lower the barrier to creating high-quality interfaces. He advises professionals to adapt by deeply integrating AI into their expertise and developing multidisciplinary skills to remain competitive. While acknowledging current limitations like occasional performance issues with complex 3D animations, the author firmly believes that tools like QoderWork Design Desk are powerful productivity enhancers that bridge the gap between design and code, making visual assets a shared responsibility across development teams.
AI袋鼠帝 • 2026-05-25 • 18,491 views
人工智能
Read original on juejin.cn →
写给年轻程序员的几点建议-2
This article offers practical advice for young programmers, spanning both career and personal development. It highlights the importance of cultivating a 'business acumen' in China's large market, encouraging readers to think innovatively, learn from successful entrepreneurs, and even consider ventures like short video creation. The author shares personal anecdotes, such as a sister's success in sales and the value of pursuing technical expertise relentlessly if one lacks a strong family background. On a personal note, the article advises prioritizing starting a family for long-term support and adopting a disciplined saving habit, suggesting a 'save before spend' approach and establishing a dream fund.
For frontend developers, the article emphasizes understanding the company's overall business operations—from deployment to monitoring—as this becomes crucial for leadership roles after age 35. It recommends enrolling in structured courses to build a systematic knowledge base, contrasting it with fragmented online content, and provides an example of how guiding interview conversations to one's areas of expertise can lead to success. The author also stresses the importance of mental well-being, advising against staying in negative work environments, and highlights the power of 'willpower' and strong aspirations, even acknowledging the role of 'destiny' in career paths.
Finally, the article suggests practicing simple algorithms to hone a programmer's mindset and thoroughly understanding fundamental concepts like `Promise` by implementing them. It concludes with the philosophy of 'doing your best and leaving the rest to fate,' encouraging effort without excessive self-blame for uncontrollable outcomes.
程序员大卫 • 2026-03-03 • 14,117 views
前端
Read original on juejin.cn →
最近公司烧token比招员工还贵，老板干懵了...
The article highlights a surprising and costly trend in the AI era: companies are finding their AI token consumption costs are rapidly exceeding their employee salaries, leaving executives bewildered. This isn't a joke; Uber's CTO revealed that their AI budget for all of 2026 was depleted in just the first four months of this year, with individual employees costing $500-$2000 monthly in AI usage. Similarly, an Nvidia VP noted that compute costs now far surpass employee costs for their team, a significant statement coming from a hardware company specializing in compute.
This phenomenon is counter-intuitive, as the software industry has long enjoyed near-zero marginal costs, making digital products cheaper with increased use. However, AI compute behaves more like a high-consumption industrial raw material than infinitely replicable code. This is largely due to the rise of AI Agents, which move beyond simple Q&A to autonomously break down tasks, perform deep reasoning, repeatedly call tools, and self-correct, leading to an exponential increase in token consumption far beyond linear estimates.
The core issue isn't just the inherent cost of tokens, but rather how they are being utilized. Many companies, in their eagerness to embrace AI, are inefficiently burning tokens, sometimes even using token consumption as a misguided KPI (dubbed "Tokenmaxxing"). Observations from Uber show vast disparities in token usage among employees for the same tasks, with no direct correlation between token consumption and actual product value. This suggests that while companies are investing in advanced AI models, their workflows and management approaches remain outdated, leading to expensive compute costs covering existing compatibility issues. The article concludes that this "token more expensive than people" situation is a growing pain of AI transformation, forcing companies to move beyond simply adopting tools to fundamentally redesign processes, ensuring every unit of compute is precisely used for irreplaceable value creation.
CodeSheep • 2026-06-02 • 17,217 views
前端 后端 程序员
Read original on juejin.cn →
为什么有的人偏爱 Mac
The author, a long-time Windows user (15 years), recently transitioned to a top-spec M4 Mac at a new company. Initially, the switch was challenging due to unfamiliar key layouts and trackpad behavior. However, after three months, the author has identified several compelling reasons why many professionals, especially programmers, favor Macs, alongside some notable drawbacks.
Key advantages highlighted include the ability to remap keys for a familiar experience, the superior pressure-sensing trackpad (despite initial functional limitations), and a more optimized software ecosystem where IDEs and development tools like Codex perform significantly better than on Windows, which often suffers from lag even on high-spec PCs. The author also praises Time Machine for its robust and automatic full-system backup capabilities, the elegant animations and unified aesthetic design that make the user experience visually appealing, and the top-tier hardware, from the programmer-friendly keyboard layout (without a numeric keypad on 16-inch models) to the display and audio quality.
Further benefits include the ARM chip's ability to run mobile applications directly and superior third-party software integration, exemplified by seamless phone unlocking via Mac's fingerprint sensor, a feature far more convenient than its Windows counterpart. However, the author acknowledges two significant drawbacks: the high cost of Mac hardware, particularly for higher specifications, and the considerable habit-forming cost for Windows users due to entirely different shortcut systems and software installation methods, requiring a period of adaptation.
小兵张健 • 2026-04-04 • 27,284 views
后端 程序员 人工智能
Read original on juejin.cn →
别再裸用 Claude Code 了！32 个亲测Skills + 8 个 MCP，开发效率直接拉满！
This article provides a comprehensive guide to significantly boosting development efficiency with Claude Code by leveraging its 'Skills' and 'MCP Servers.' It addresses common developer pain points such as Claude Code's limited understanding of requirements, generating unusable UI, inability to interact with local files or browsers, and the loss of context across sessions.
The core distinction between Skills and MCP Servers is clearly explained: Skills are encapsulated prompts and standardized workflows that make Claude 'smarter' and more specialized in specific domains, while MCP Servers are actual tool capabilities running locally, enabling Claude to 'perform actions' like accessing local files, browsers, or external APIs. A key advantage is that most of these capabilities are automatically triggered, eliminating the need for manual invocation.
The guide details 32 hand-picked Skills, categorized by development scenario (frontend, documentation, architecture, memory, testing, efficiency, security, and custom development), providing installation commands, core capabilities, trigger scenarios, and real-world insights for each. It also offers convenient one-click installation scripts tailored for beginners, frontend developers, and full-stack developers. Furthermore, it covers 8 tested MCP Servers, including a neural memory system, browser automation (Playwright), file system access, sequential thinking, web content scraping, and integration with Figma for both reading and editing. Each MCP entry includes installation prerequisites, full configuration, core capabilities, and important notes, along with a summary of API key requirements. The article concludes with a crucial pitfall avoidance guide for both Skills and MCPs, covering common installation/configuration issues and essential security warnings, a Q&A section for beginners, and a recommended installation order to help users get started effectively.
蝎子莱莱爱打怪 • 2026-03-23 • 72,711 views
后端 Claude Java
Read original on juejin.cn →
面试官问我：“AI 写代码比你快 100 倍，你的价值在哪？”
This article addresses a common anxiety among developers: the rise of AI in coding. The author recounts an interview where a technical director challenged him on the core value of programmers when AI can code faster and with fewer errors. The author argues that while AI makes the 'craft of coding' cheaper, it doesn't simplify 'building software.' Human value is shifting from mere execution to higher-level cognitive tasks like demand insight, architectural trade-offs, and ultimate accountability for system failures, which AI cannot provide.
The article then outlines five crucial skills for developers to cultivate in the next five years to build a 'moat' around their careers. These include evolving from 'package callers' to architects, mastering prompt engineering to leverage AI effectively, developing keen code review abilities to 'clean up' AI-generated code, cultivating a business and product mindset, and focusing on using AI to solve real-world business problems rather than just discussing theory. The author emphasizes that true innovation lies in applying existing AI capabilities to address long-standing pain points.
Finally, the article offers tailored breakthrough strategies for developers at different career stages, from newcomers learning design patterns to senior veterans focusing on technical strategy and cross-functional experts leveraging AI for vertical innovation. The core message is that AI is not an enemy but a powerful tool. Programmers' future lies in a 'leap in thinking' – commanding AI as a tireless assistant to tackle high-value, complex problems that require human insight, decision-making, and ultimate responsibility.
雨夜寻晴天 • 2026-04-16 • 35,429 views
前端 AI编程 面试
Read original on juejin.cn →
全面封禁 Cursor！又一家大厂出手了
Kuaishou's R&D department has recently restricted the use of third-party AI programming tools like Cursor, rendering them unusable for employees. This move follows a similar decision by ByteDance in May, which began disabling Cursor and Windsurf to promote its self-developed programming assistant, Trae. ByteDance cited data security concerns, as employees purchased tools through personal accounts, leading to data storage outside company control, and compliance risks due to some models not officially serving the Chinese market. This caused significant backlash among ByteDance programmers, who found Trae to be inferior to Cursor at the time.
The trend of restricting third-party AI tools extends beyond Chinese tech giants. Amazon, in November, mandated the prioritization of its in-house AI tool, Kiro, and ceased supporting new third-party options. Similarly, Microsoft in September banned DeepSeek-related applications, pushing its own GitHub Copilot. The primary reasons behind these corporate bans are data security—preventing sensitive company code from being sent to external cloud models or used for training—and concerns over the inconsistent quality of AI-generated code, which can lead to critical online incidents if not thoroughly reviewed.
Despite these corporate restrictions, AI programming is an undeniable and rapidly accelerating industry trend. Data from Tencent's '2025 R&D Big Data Report' indicates that over 90% of its programmers use AI assistants, and by 2025, 50% of all new code within the company is expected to be AI-assisted. The author emphasizes that even if companies restrict certain external AI tools, developers must actively learn and utilize the latest AI programming tools to maintain their competitiveness in a fast-evolving technological landscape.
程序员鱼皮 • 2025-12-16 • 192,299 views
Cursor AI编程 程序员
Read original on juejin.cn →
又是梁文锋，有点猛啊。
This article delves into the remarkable journey of Liang Wenfen, the founder of DeepSeek, a prominent name in the AI industry. It begins by highlighting his exceptional academic prowess from an early age, notably his astounding 806 score on the Gaokao (National College Entrance Examination) over two decades ago, which earned him the title of top scorer in his region. Despite his current global recognition, Liang Wenfen maintains a remarkably low profile, rarely appearing in public or engaging in extensive marketing for DeepSeek, sparking curiosity about his path to becoming a top AI entrepreneur.
The article systematically traces Liang Wenfen's academic and entrepreneurial trajectory. Born into a family of teachers, his innate talent for mathematics led him to excel academically, culminating in his enrollment at Zhejiang University for both his undergraduate and master's studies, specializing in Electronic Information Engineering and Information and Communication Engineering, respectively. His entrepreneurial spirit emerged during his university years, driven by a keen interest in financial markets and the potential of quantitative investment powered by mathematics and AI. This led him to co-found Hangzhou Jacobi Investment Management Co. in 2013, and then High-Flyer Quant (幻方量化) in 2015, which rapidly grew to manage assets worth hundreds of billions, establishing him as a 'low-key quantitative king' by 2021.
Building on his success in quantitative finance, Liang Wenfen pivoted to the burgeoning field of Artificial General Intelligence (AGI), founding DeepSeek in 2023. DeepSeek quickly made waves with its innovative, open-source large language models, notably DeepSeek-R1, which challenged the dominance of international tech giants and promoted the democratization and accessibility of AI. The article concludes by sharing 50 insightful thoughts from Liang Wenfen on DeepSeek's mission, AI industry trends, technology, innovation, recruitment, and open source, offering valuable lessons for aspiring entrepreneurs and technologists.
CodeSheep • 2026-06-16 • 13,418 views
前端 后端 程序员
Read original on juejin.cn →
Codex 穷鬼大救星
This article introduces `codex_with_cc`, an open-source workflow designed to make powerful AI models like Codex more cost-effective for developers. It addresses the problem of high token consumption when Codex handles complex, context-heavy tasks by itself, proposing a multi-agent delegation system to significantly reduce operational costs.
The core of the workflow positions Codex as the "leader" or "architect," responsible for high-level tasks such as task decomposition, delegation, review, and final acceptance. Meanwhile, "sub-agents," powered by Claude Code CLI and backed by the low-cost DeepSeek API (especially leveraging its high cache hit rate), are assigned the heavy-lifting tasks. These include deep code analysis, extensive refactoring, debugging, and testing. This division of labor prevents the main Codex thread from being bogged down by "dirty work" and excessive context, allowing it to focus on critical judgment and decision-making, while DeepSeek handles the token-intensive, repetitive tasks cheaply.
Beyond simple prompt-based multi-agent systems, `codex_with_cc` emphasizes engineering robustness. It incorporates features like a Claude session reuse pool to minimize redundant context loading, task fingerprinting, lease management for parallel workers, and comprehensive audit artifacts. This structured approach ensures traceability, efficiency, and accountability in multi-agent collaboration, making it ideal for large-scale code exploration, multi-file implementations, solution brainstorming, and separating implementation from code review, ultimately reducing token costs and enhancing the development experience for complex projects.
小呆呆666 • 2026-05-04 • 31,274 views
前端 后端 人工智能
Read original on juejin.cn →
程序员副业 | 2026年4月复盘
The author, a programmer with a decade of experience who transitioned to entrepreneurship 900 days ago, presents his April 2026 monthly review. A significant shift in his business direction occurred, moving away from the ComfyUI niche, which he pursued last year, towards AI programming and AI agents, particularly focusing on tools like Claude Code and OpenCLaw (Xiaolongxia). This strategic pivot led to new engagements, including multiple OpenCLaw sharing sessions and a month-long corporate training and delivery project for AI agents.
This shift meant a temporary halt in ComfyUI-related content creation and income, which was successfully offset by earnings from training and corporate projects. A major highlight of April was the development and launch of a new software called "Storybound." This tool automates the creation of video content for book recommendation e-commerce on WeChat Channels, handling script rewriting, image generation, voiceovers, and video synthesis. Using Storybound, the author achieved impressive results, gaining 90,000 new followers and 15 million video views in April, generating over 20,000 RMB in GMV and more than 10,000 RMB in commission.
The development of Storybound, primarily using Claude Code, rekindled the author's passion for coding and proved to be a profitable venture, with plans for further iteration and expansion. He also clarified his role in an offline company, now acting as a collaborator for specific projects rather than a full-time participant, focusing instead on his independent ventures. The author's community maintained a healthy 60%+ renewal rate, fostering collaborations on various AI-driven projects. Looking ahead, the author emphasizes the importance of "compound interest" activities and the efficiency gains from AI tools, while also addressing personal challenges like sleep and time management, with a clear plan for health and software refinement in May.
嘟嘟MD • 2026-05-03 • 19,102 views
后端 创业
Read original on juejin.cn →
我放弃了 Vue/React，选择自研框架
The author discusses the current "framework-is-king" state of frontend development, dominated by Vue and React. While these frameworks offer vast ecosystems, they impose significant cognitive burdens, high upgrade costs, and ecosystem lock-in in commercial projects. The critical turning point for the author was a project requiring a high-performance, customizable visualization editor, a need that existing frameworks like Vue, React, and Solid couldn't fully meet, specifically regarding fluid rendering with high-frequency state updates, replaceable rendering backends, and fine-grained control over the rendering process.
The article elaborates on the shortcomings of mainstream frameworks. Vue's built-in components are described as "black boxes," limiting customization. React suffers from the complexity of Hooks' dependency arrays and the inherent performance overhead of its virtual DOM in high-frequency update scenarios. Signal-driven frameworks like Solid and Svelte, while performant, introduce issues with opaque compiled output and restricted runtime flexibility due to extensive compile-time optimizations.
Driven by these challenges, the author developed Vitarx, a framework built on three core principles: signal-level precise updates (eliminating diffing overhead), a runtime view tree for dynamic operations, and "everything is a component" for full transparency and extensibility. Vitarx features a layered architecture and a unique reactive system using a doubly linked list for efficient dependency tracking. The article also details solutions to key development challenges, including memory management for reactive systems, efficient view tree updates via ViewSwitchTransactions, a custom JSX compilation pipeline, and robust server-side rendering. Performance benchmarks show Vitarx outperforming Vue and React in specific update scenarios, with smaller bundle sizes. The author concludes by reflecting on the deep understanding and control gained, advising others to have clear goals and iterate continuously when considering self-development.
下家 • 2026-06-25 • 50 views
前端 前端框架
Read original on juejin.cn →
面试官坏笑：“你用 AI 编程一年了，怎么保证 Claude Code 写出来的代码是对的？”我：“直接上 Claude Fable 5 啊！”
This article addresses two common questions about Claude Code: how to use it and, more importantly, how to ensure the correctness of the code it generates. The author, Er Ge, condenses his experience into six practical cases, each demonstrating specific prompts, execution processes, and results to help users understand how to leverage AI for reliable coding.
The core strategies revolve around providing robust context and implementing deterministic checks. Key techniques include using `CLAUDE.md` as a project-level instruction manual to define tech stacks, coding standards, and forbidden commands, ensuring the AI operates within defined boundaries. The article also highlights the power of 'Skills' – specialized knowledge packages that transform the AI into an expert for specific domains, such as using a `web-access` skill for in-depth competitive research. Furthermore, 'Hooks' are introduced as code-level intercepts that enforce deterministic rules, running outside the probabilistic model to perform tasks like compilation checks or security intercepts, ensuring code quality and adherence to standards.
Beyond these, the article emphasizes efficient context management, treating it as the AI's most valuable resource. It advises against installing too many global skills, breaking complex tasks into smaller, independent sessions, and using `CLAUDE.md` for persistent project information. The author also shares principles for crafting effective prompts, stressing the inclusion of the 'object of operation,' 'action,' and 'verification method' for higher accuracy. Finally, the article showcases the newly released Claude Fable 5 model's impressive capabilities, demonstrating its use in generating a promotional PPT with top-tier aesthetics, reinforcing the idea that success in AI programming lies not just in the model's power, but in how effectively one integrates its capabilities with engineering tools to create a repeatable and verifiable workflow.
沉默王二 • 2026-06-14 • 4,630 views
Claude Agent AI编程
Read original on juejin.cn →
苦撑13年，创始人离职出走，拉勾终究还是倒下了…
Lagou, once a prominent internet recruitment platform, has reportedly collapsed after 13 years of operation. The author, an early user, expresses regret, noting the platform's current state: an accessible but largely unmaintained website with outdated job postings and interview reviews, non-functional login, and absent mobile apps. Its official social media accounts haven't been updated for over a year, and its corporate registration is not up-to-date, signaling a quiet end to its journey.
Founded in 2013 by Xu Dandan, Lagou quickly rose to prominence by focusing exclusively on internet industry recruitment. It differentiated itself from established players like Zhaopin and 51job with clear salary ranges, streamlined application processes, and a targeted approach. This innovative strategy attracted millions of job seekers and thousands of internet companies, leading to significant funding rounds and cementing its status as a unicorn in its niche.
However, a turning point arrived in 2017 when 51job acquired a 60% controlling stake. Despite assurances of independent operation, Lagou gradually lost its initial agility and sharp focus. Strategic missteps, such as venturing into IT education and headhunting, coupled with the rise of competitors like BOSS Zhipin and a broader economic downturn (slowdown in internet growth, the pandemic), severely weakened its core business. The founder, Xu Dandan, resigned in 2023, leaving 51job in full control, marking the definitive end of Lagou's era. The article concludes by reflecting on how Lagou's initial success was tied to the mobile internet boom, highlighting the importance of continuous innovation beyond merely riding an "era's dividend."
CodeSheep • 2026-05-29 • 17,766 views
前端 后端 程序员
Read original on juejin.cn →
我用3个周末重构了公司的前端项目，老板没发现，但同事都来找我要代码了
This article details the author's experience of refactoring a 3-year-old Vue2 backend system to Vue3 + TS + Vite over three weekends. The original project suffered from inconsistent coding styles due to four previous frontend developers, lacked TypeScript, component library standards, and was riddled with `console.log` statements. The author undertook this ambitious refactor during a business lull, emphasizing that it didn't disrupt normal development cycles.
The refactoring process was broken down into three main phases. The first weekend focused on migrating the core framework, tackling the painful transition from Options API to Composition API, upgrading Vue Router from v3 to v4, and migrating Vuex to Pinia, which offered superior TypeScript support. A key lesson learned was the importance of meticulously checking migration guides after encountering an issue with `v-model`'s changed behavior. The second weekend was dedicated to integrating TypeScript, addressing common pitfalls like `defineComponent` props types, `ref` generic inference, and creating custom `.d.ts` files for older third-party libraries. The author highlighted the immense satisfaction of finally having proper IDE auto-completion.
The final weekend concentrated on engineering optimizations to improve maintainability. This included setting up ESLint + Prettier + Husky for automated code formatting, implementing `unplugin-vue-components` for automatic component imports, and centralizing environment variable management using `.env` files. The author concludes by sharing that the true value of refactoring isn't about using the latest tech, but about enabling even the least experienced team members to write quality code through robust tooling and practices. The success was evident when colleagues from other teams sought out his configurations for their own Vue3 upgrades.
柯克七七 • 2026-06-09 • 5,787 views
前端
Read original on juejin.cn →
🚀 2026 年 5 月 GitHub 十大热门项目排行榜 🔥
This article presents the top 10 GitHub open-source projects for May 2026, carefully selected from 17 monthly candidates for their significant long-term value. These projects span diverse and critical areas, including engineering skill libraries, code knowledge graphs and memory, multi-agent orchestration, industry-specific plugin templates, AI engineering education, and fully automated short video production. A central theme uniting these selections is the open-source community's ongoing effort to evolve AI agents from simple, single-turn conversational tools into robust, reusable, collaborative, and deliverable productivity solutions.
The featured projects highlight four key trends shaping the AI and development landscape. Firstly, there's a strong emphasis on "skill and plugin assetization," where common engineering actions, academic research workflows, and industry-specific roles are codified into reusable skills or plugins. Secondly, "context engineering" is gaining traction, with projects focused on reducing agent costs by enhancing retrieval, understanding, and cross-session memory. Thirdly, advancements in "orchestration and collaboration" are evident, with platforms enabling multi-agent coordination to bridge the gap from individual CLI usage to team-based workflows. Finally, the importance of "content and engineering best practices" is underscored, covering AI-driven content automation alongside the crucial need for engineering discipline beyond mere tool utilization.
一点一木 • 2026-06-01 • 24,170 views
GitHub 人工智能 AI编程
Read original on juejin.cn →
面试官问你：“AI 能写 80% 的代码了，公司为什么还需要你？”
This article addresses a prevalent and challenging question in modern tech interviews: "AI can write 80% of the code, so why does the company still need you?" The author highlights that traditional answers, such as AI's code quality being poor, its inability to understand business needs, or humans solely performing code reviews, are no longer sufficient in 2026. These responses often lead interviewers to question the candidate's value or equate their skill level with AI.
The author proposes a three-layered framework for a compelling answer, emphasizing that the question assesses one's self-worth rather than their stance on AI. First, at the execution layer, AI writes code, but humans make critical decisions, anticipating unstated problems and potential pitfalls that AI overlooks. Second, at the architecture layer, AI can generate individual components, but humans design entire systems, making strategic architectural choices based on business context, team capabilities, and future scalability. The article stresses that while AI lowers the barrier to writing code, it raises the bar for making correct decisions, as wrong choices can rapidly accumulate technical debt.
Finally, at the responsibility layer, the author points out that AI cannot be held accountable for production incidents. Companies pay for human judgment, communication skills, and a sense of responsibility to diagnose, resolve, and prevent future issues. The article concludes by identifying core competencies that become more valuable as AI advances: system design, deep business understanding, cross-team collaboration, and the ability to handle and resolve live production issues. Providing a concrete, real-world example is also crucial to illustrate these points effectively.
kyriewen • 2026-06-25 • 432 views
前端 JavaScript 面试
Read original on juejin.cn →
白嫖DeepSeek V4 Pro！免费无限用，还能接入Claude-Code
This article provides a straightforward, 5-step guide to accessing DeepSeek V4 Pro's free tier directly from your local development tools, eliminating the need for payments or complex setups. By leveraging the ZenMux platform for API key management and local API forwarding tools like cc switch or Cherry Studio, users can tap into powerful AI models for various tasks, including complex reasoning, coding, and long-form writing, all at no cost. The process is designed to be quick and efficient, making advanced AI capabilities accessible to everyone.
The core steps involve registering for a ZenMux account, generating an API Key, and crucially, restricting this key to specific free models such as `deepseek/deepseek-v4-flash-free` and `deepseek/deepseek-v4-pro-free`. Once the key is obtained and restricted, users configure their chosen local tool (e.g., Cherry Studio or cc switch) with the ZenMux API key and the appropriate API endpoint. The article provides detailed instructions and screenshots for both Cherry Studio and cc switch configurations, ensuring a smooth setup process.
While this free offering is highly valuable, the author advises users to be mindful of a few considerations. Free models might experience slower responses or temporary unavailability during peak hours, and their capabilities, such as context length and reasoning depth, may be slightly limited compared to paid versions. Additionally, users are strongly cautioned about API key security, emphasizing the importance of never exposing keys in public repositories or sharing them carelessly. Despite these minor caveats, the free DeepSeek V4 Pro access is presented as an exceptionally cost-effective solution for both personal daily use and developer API testing.
神奇小汤圆 • 2026-06-14 • 5,822 views
后端
Read original on juejin.cn →