Shopify’s CEO threatened to ban Claude Code. Anthropic had already closed the feature request.
The New Stack Amanda Caswell
Shopify’s CEO says he may ban Claude Code over how it reads repo instructions. The fight is really about AI agents staying in sync inside huge codebases.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Shopify CEO Tobi Lütke is talking about banning Claude Code at work, and the target isn’t Anthropic’s model quality. It’s the way the tool handles project instructions. On Tuesday, Lütke posted on X that he was thinking about blocking Claude Code at Shopify until it “read AGENTS.md and .agents/skills etc.” instead of relying on its own file format.
That sounds tiny until you remember what Shopify looks like from the inside. The company has thousands of developers working in a single massive monorepo, and not every team uses the same AI coding tool. In that setup, instruction files are doing real coordination work. If one agent sees a different set of rules than another, the result is split-brain behavior: different tools, different context, different output.
Lütke later said the company has already built automation to smooth over the mismatch. But he doesn’t seem impressed by the fact that the workaround exists. His complaint is that engineers should not have to pay what he called a “stupid complexity tax” just because the tools don’t agree on a common instruction file.
This is where AGENTS.md and CLAUDE.md matter. Both are meant to give coding agents persistent guidance about a repository: build commands, testing rules, coding conventions, and the like. OpenAI introduced AGENTS.md in August 2025, and by December said it had spread to more than 60,000 open-source projects and agent frameworks, with support from tools including Codex, Cursor, Gemini CLI, GitHub Copilot, Jules and VS Code. OpenAI later handed the format to the Agentic AI Foundation under the Linux Foundation.
Claude Code, though, uses CLAUDE.md and does not natively read AGENTS.md. Anthropic has offered workarounds, such as importing AGENTS.md from CLAUDE.md or using symbolic links, but those fixes still leave teams maintaining the same instructions in multiple places. In a big repository, that is exactly the kind of housekeeping that turns AI help into platform-team busywork. Anthropic has already closed a request for recursive AGENTS.md discovery as “not planned,” which tells you enough about how much urgency it sees here.
The broader point is simple: once companies let several coding agents into the same codebase, file-format politics stop being trivia. At Shopify’s scale, instruction drift is not a corner-case annoyance. It is the sort of thing that quietly multiplies until somebody threatens to ban a tool over a Markdown file.
My take — AI-written commentary, not fact-checked reporting
This is the real AI enterprise story: not model magic, but standards, plumbing, and who gets stuck cleaning up the mess. Anthropic can keep saying “not planned,” but platform teams are the ones paying for that decision in duplicated instructions and avoidable drift. The industry keeps selling autonomy; the bill arrives as coordination.
Read more about this at: The New Stack
Related stories
Anthropic Released Claude Commerce Agents: An Apache-2.0 Blueprint for Shopping and Merchant Agents Across Retail, Travel, Telecom and Entertainment
MarkTechPost · 13 hours ago ·
31
A Fireside Chat with Cat and Thariq from the Claude Code team
Simon Willison's Weblog · 1 month ago ·
47