Shopify's Claude Code Ban Threat: The AGENTS.md Fight (2026)
Shopify's CEO threatened to ban Claude Code over one missing file: AGENTS.md. Here is what actually happened in August 2026, why Anthropic finally responded, and what it teaches anyone who runs real work through AI agents.
⚡ TL;DR
In late August 2026, Shopify CEO Tobi Lütke publicly said he was considering banning Claude Code across Shopify because the tool reads its own CLAUDE.md configuration file but ignores AGENTS.md, the vendor-neutral standard most competing tools already support. After nearly a year of declining the request, Anthropic responded through a Claude Code team member who said the company is working on making the tool easier to customize, including simpler use of AGENTS.md.
That is the whole story in two sentences. The interesting part is what it reveals about lock-in risk for anyone who builds their daily workflow on top of one AI vendor's file format.
📰 What Happened
On August 25, 2026, Shopify CEO Tobi Lütke posted on X that he was thinking about banning Claude Code company-wide until Anthropic changed how the tool reads repository instructions. The complaint, reported by BigGo Finance and The New Stack, centers on a specific technical behavior: Claude Code defaults to reading a proprietary CLAUDE.md file, while ignoring AGENTS.md, an open configuration standard that tools like OpenAI's Codex, Cursor, Amp, and GitHub Copilot already read.
Lütke described the situation as a complexity tax for teams that manage AI coding agents across large codebases. When a company runs multiple AI tools, and Shopify gives its employees several, each repository needs duplicate instruction files or the tools behave inconsistently. Developers call this a split brain problem.
This was not a new request. The main GitHub feature request asking Claude Code to support AGENTS.md was filed in August 2025 and collected close to 5,000 upvotes, making it one of the most supported requests in the Claude Code community. Anthropic had repeatedly marked the request as not planned.
After Lütke's post gained traction, Claude Code team member Thariq Shihipar responded publicly. He said Anthropic is working on making Claude Code more hackable, which includes making AGENTS.md easier to use and opening up broader customization of the system prompt.
What AGENTS.md Actually Is
AGENTS.md is a plain markdown file that lives in a code repository and tells AI coding agents how to work in that project: build commands, test procedures, style rules, and security guidelines. Its entire value is that it is tool-neutral. Write it once, and any agent that supports the standard follows the same playbook.
🎯 Why It Matters
This looks like a developer-tooling squabble, but the stakes are bigger. Instruction files like AGENTS.md and CLAUDE.md are becoming the operating manuals of AI-driven work. Whoever controls the format controls a piece of your workflow.
For a solo operator, the math is different than it is for Shopify, but the risk is the same shape. Shopify can threaten a ban and get a public response from Anthropic within days. You cannot. If your entire agent setup lives in one vendor's proprietary format, you inherit every one of that vendor's decisions with zero leverage.
The support landscape as of August 2026 makes the divergence obvious.
| Tool | Reads AGENTS.md | Primary Config File |
|---|---|---|
| Claude Code (Anthropic) | No, by default | CLAUDE.md |
| Codex (OpenAI) | Yes | AGENTS.md |
| Cursor | Yes | AGENTS.md |
| Amp | Yes | AGENTS.md |
| GitHub Copilot | Yes | AGENTS.md |
💬 Our Take
We run this one-person company on AI agents, and a large share of that work flows through Claude Code. So we have skin in this game, and our honest read is this: both sides are right, and that is exactly why the outcome matters.
Anthropic's position made a certain kind of sense. CLAUDE.md and the skills system are tightly designed, and a tool that deeply integrates its own format can deliver a better experience than one that treats configuration as a lowest common denominator. But a nearly year-old feature request with close to 5,000 upvotes marked as not planned is not a design philosophy. It is a bet that users will tolerate lock-in. Lütke just showed the ceiling on that bet: the moment a customer is big enough, the answer flips from not planned to we are working on it.
The lesson we take as a solo shop is not to abandon Claude Code. It is to stop treating any vendor's config file as the source of truth. Our project instructions should live in a neutral format we own, with thin adapters per tool. That is a one-hour refactor today and an insurance policy against every future dispute like this one.
The second lesson is about signals. Public pressure from one CEO moved Anthropic further in one week than a year of GitHub upvotes did. As small operators we cannot generate that pressure alone, which is precisely why open standards matter more to us than to Shopify. Standards are how small players inherit the leverage of big ones.
🛠️ What to Do Today
You do not need to wait for Anthropic to ship anything. Two moves protect you right now.
First, make one canonical instruction file and reference it everywhere. Keep your real project rules in AGENTS.md, then make CLAUDE.md point to it. A symlink works, and so does a one-line file that says to read AGENTS.md and follow it. Claude Code follows that instruction today.
Second, audit your setup for vendor-specific sprawl. If your workflow rules are scattered across CLAUDE.md files, custom skill folders, and tool-specific settings, consolidate the parts that describe your business logic into the neutral file. Keep only tool-specific mechanics in tool-specific files.
- ✔Create a single AGENTS.md with your build, test, and workflow rules
- ✔Add a CLAUDE.md that simply instructs the agent to read AGENTS.md
- ✔Move business logic out of tool-specific config into the neutral file
- ✔Re-test each AI tool you use against the consolidated setup
- ✔Note which of your tools support AGENTS.md natively, and watch Anthropic's changelog for the promised update
# CLAUDE.md (bridge file) Read AGENTS.md in this repository and follow all instructions in it. Treat AGENTS.md as the source of truth for build commands, tests, and style rules.
❓ Frequently Asked Questions
Did Shopify actually ban Claude Code?
No. As of late August 2026, Tobi Lütke said he was considering a ban unless Anthropic changed the file-reading behavior. It was a public pressure move, not an executed policy, and Anthropic responded within days.
What is the difference between AGENTS.md and CLAUDE.md?
Both are markdown files that give AI coding agents instructions about a repository. AGENTS.md is an open, tool-neutral standard read by Codex, Cursor, Amp, and GitHub Copilot. CLAUDE.md is Anthropic's own format, and Claude Code reads only it by default.
Will Claude Code support AGENTS.md now?
Anthropic has not announced a shipped feature or a date. Claude Code team member Thariq Shihipar said the team is working on making the tool more hackable, including easier use of AGENTS.md and broader system-prompt customization. Until it ships, a bridge CLAUDE.md that points to AGENTS.md is the practical workaround.
Why did Anthropic refuse the request for so long?
Anthropic never published a detailed rationale. The main GitHub request, filed in August 2025 with close to 5,000 upvotes, was marked not planned. Observers generally read it as a product strategy choice favoring Anthropic's own configuration ecosystem over the cross-vendor standard.
🏁 Final Thoughts
The Shopify versus Anthropic dispute ends with a familiar pattern: a big customer applied public pressure, and a not planned label turned into we are working on it. The verified facts are simple. Lütke threatened a ban on August 25, 2026, the AGENTS.md request had sat for nearly a year with close to 5,000 upvotes, and Anthropic finally signaled movement. The takeaway for solo operators is to own your instructions in a neutral format and let every tool adapt to you, not the other way around. We publish field notes like this from a one-person company run by AI agents several times a week. Subscribe to catch the next one, and tell us in the comments: do you keep your agent rules in AGENTS.md, CLAUDE.md, or both? Source: BigGo Finance, https://finance.biggo.com/news/e2596dfa-cc76-4b50-b487-c54b2caffc3b. Related reading on this blog: our earlier post on choosing an AI coding agent as a solo founder.
Last updated: August 27, 2026 · Keyword: Shopify Claude Code AGENTS.md · Agents at Work
Comments
Post a Comment