Shopify CEO怒喷Claude Code！

Shopify CEO怒喷Claude Code！一份文件引发的“封杀令”

朋友们，AI圈又吵起来了！这次不是比谁家模型更强，而是一份小小的配置文件，直接让全球电商巨头Shopify的CEO Tobi Lütke 公开威胁：要么支持行业标准，要么在Shopify内部全面封杀Claude Code！

🤯 一份文件，怎么就成了“复杂度税”？

事情的导火索，是Claude Code拒绝读取行业通用的 AGENTS.md 格式，固执地只认自家的 CLAUDE.md。

你可能觉得，换个文件名而已，至于吗？但在Shopify这种有数千名开发者共同维护一个庞大代码库的公司里，问题就大了。不同AI工具读的指令文件不一样，有的开发者用了“完整版”AI，有的却像“被切掉一块脑子”一样缺少关键指令。为了填这个坑，团队不得不用自动化脚本去修补，CEO直言这是一笔“愚蠢的复杂度税”。

😡 开发者苦“特立独行”久矣！

其实，开发者们早就受够了。近一年来，要求Claude Code支持AGENTS.md的呼声一直很高，GitHub上相关功能请求获近5000个赞，却接连被官方以“不予修复”关闭。

更绝的是，为了让Claude Code读上AGENTS.md，有大神分享了一段极其复杂的提示词——要写脚本递归查找、创建符号链接、监控文件变化、设置开机自启……一套操作下来，工程量堪比部署一个小型服务。网友直呼：为了兼容，开发者被逼写了多少“无用”脚本？

🎤 Anthropic回应了，但开发者更气了

眼看舆论发酵，Anthropic终于回应：未来会提高可定制性，允许开发者更方便地使用AGENTS.md。但话锋一转，他们坚持认为不同模型需要的指令不同，拒绝像其他工具一样默认读取行业标准。

这番回应直接点燃了社区怒火。开发者质问：如果每个模型都要配专属指令，难道我要维护17份不同的文件？每当模型更新，是不是还要专门雇个团队天天改Markdown？这态度，传递出的信息就是：OpenAI不对，开源社区不对，只有我Anthropic知道最佳方案。

💎 傲慢是开发者流失的开始

技术特立独行没问题，但当你的“特立独行”开始给全球开发者增加无谓的成本时，这就不是个性，而是傲慢了。正如一位网友所说，Anthropic正在因这种文化问题，逐渐失去开发者的关注。

#ClaudeCode #Anthropic #AI编程 #开发者 #科技圈大事件 #程序员日常 #效率工具

以下是对图中文章 OCR 扫描的结果：

tobilutke口…

@tobiSo many people telling me tricks for dealing with this-I know them all.But agents and claude files are recursively applied through the tree. Withthousands of developed in a mono repo, it just does happen that onedirectory is missing one of the two files and this means that a subset ofdevs work with lobotomy. We fix this with automation but it's a stupidcomplexity tax that shouldn't have to be paid.很多人都跟我分享过解决这个问题的技巧—我全都知道。但是代理文件和Claude 文件是递归地应用到整个代码树中的。在一个单体仓库里开发了成千上万个项目，难免会遇到某个目录缺少这两个文件中的一个的情况，这意味着一部分开发人员的工作会变得非常糟糕。我们可以通过自动化来解决这个问题，但这纯粹是不必要的复杂性代价。

1:48 AM • Aug 26,2026•45.1K Views

tobilutke口⋯@tobiI'm thinking about banning Claude code at Shopify until they changetheir mind and read AGENTS.md and .agents/skills etc.Insisting on only reading CLAUDE.md sometimes leads to split brainproblems when different team members use different tools. Justunnecessary.Last edited 10:34 PM • Aug 25, 2026•1M Views

lan Nuttall &口⋯@iannuttallUntil Claude supports this just use the prompt below to automaticallycreate a symlink for CLAUDE.md to AGENTS.md in your dev folder（s）—Set up an auto-symlink system on my Mac so l only ever maintainAGENTS.md, never CLAUDE.md.First, ask me which directory （or directories） contain my code projects.Use that as ROOT throughout. Do not assume a path.Requirements：1. Create ~/bin/link-agents.sh—aone-shot script that recursively findsevery AGENTS.md （case-insensitive） under ROOT （accept it as anargument, default to the path I gave you）， skipping node_modules, git，and vendor, and creates a relative symlink CLAUDE.md -> AGENTS.mdin the same folder if no CLAUDE.md already exists. Run it once againstROOT.2. Create ~/bin/watch-agents.sh — uses fswatch （install via Homebrewif missing） to recursively watch ROOT and create the symlink whenever anew AGENTS.md appears.3. Create a launchd agent at~/Library/LaunchAgents/com.linkagents.plist that runs the watcher atlogin with KeepAlive and ROOT passed as an argument, PATH including/opt/homebrew/bin. Load it.4. Add CLAUDE.md to my global gitignore （~/.gitignore_global, setcore.excludesfile if not already set） so the symlinks are never committed.5. Make both scripts executable, then verify: create a temp folder with anAGENTS.md under ROOT, confirm CLAUDE.md appears within a fewseconds, then clean up the temp folder.

0…

Srkprasad@Srkp1729Yes, I too attempted many fixes to make Claude Code read AGENTS.md，but it simply does not respond. Eventually, Iswitched to cursor.4:18 AM •Aug 26,2026•555 Views

Thariq0…

@trq212Hi Tobi, thanks for the feedback! We are working on making Claude Codemore hackable, which will include being able to easily use Agents.MD ormake other system prompt modifications. I'll share more here when it'sready to roll out.Our reasoning for this was that we don't think that model families areinterchangeable and the system prompt can have a big impact onperformance.

Anthropic is again refusing to follow industry standards like AGENTS.md，and insists that CLAUDE.md needs to be specialnow here's what's wrong with their stance：1. the underlying attitude from Anthropic is that they know better thaneveryone elseOpenAl is doing it wrong. the open source community is doing it wrong.everyone is doing it wrongonly we anthropic know the best way, and we will not budge2.the idea that every model needs a separate AGENTS.md is not evenself-consistentif that's the case, why do we have a CLAUDE.md？

it should be- CLAUDE_FABLE_5.md- CLAUDE_OPUS_5.md- CLAUDE_OPUS4_8.md…

every repo should pay the tax to maintain dozens of different variationsfor every single model that's ever used by any contributor, and every oneneeds to be constantly, rigorously evaluated with millions of tokensoh and every model becomes obsolete every several weeks, so justallocate a team in your company to refresh your markdown files full timeand not do any real work