Shopify CEO Considers Banning Claude Code Over AI Coding Concerns

Shopify CEO Tobi Lütke said he is considering banning Anthropic’s Claude Code from the company until the AI coding tool supports developer instruction formats beyond its native CLAUDE.md files. Lütke argued that Claude Code’s current approach can create “split brain” problems when developers using different AI coding tools rely on different instruction files across the same codebase.

The comments highlight a growing issue for companies adopting multiple AI coding agents: how to maintain consistent instructions, workflows and engineering standards when different tools use different configuration systems. The dispute is notable because Shopify itself has promoted Claude Code alongside other AI development tools, including Cursor, Gemini CLI, VS Code and OpenAI’s Codex CLI, on its developer platform.

Why Shopify Is Considering A Claude Code Ban

Lütke’s concern is centered on how Claude Code processes instructions inside a large software repository. Claude Code uses CLAUDE.md files to provide project-specific guidance to the coding agent.

According to Lütke, Shopify’s engineers also work with other agent-oriented conventions, including AGENTS.md and .agents/skills. If an AI coding agent recognizes only one of these systems, developers using different tools may receive different instructions when working on the same part of a repository.

The Core Issue

Developer Environment	Instruction Format Mentioned
Claude Code	CLAUDE.md
Other AI coding agents	AGENTS.md
Agent skills	.agents/skills
Shopify concern	Inconsistent instructions across tools
Potential consequence	Different developers receive different agent guidance

The problem becomes more complicated in a large monorepo, where instructions can be applied recursively through different directories.

Lütke later explained that Shopify has thousands of developers working within its monorepo and that a missing instruction file in one directory can cause a subset of developers to receive incomplete guidance from their AI tools. The company uses automation to mitigate the problem, but he described the additional maintenance as an unnecessary “complexity tax.”

What Is Claude Code?

Claude Code is Anthropic’s AI coding agent that operates in a developer’s coding environment and can understand a codebase, modify files, execute commands and assist with software-engineering tasks.

Unlike a conventional chatbot, coding agents are designed to work directly with repositories and development tools. That makes project-specific instructions particularly important because an agent needs to understand coding conventions, testing requirements, architectural decisions and other rules before making changes.

Anthropic has built CLAUDE.md into Claude Code as a mechanism for communicating this context.

Why Instruction Files Matter

Instruction files can tell an AI coding agent things such as:

• Which coding conventions to follow
• Which commands to run before submitting changes
• How tests should be executed
• Which files or directories require special treatment
• How the project is structured
• Which development tools or workflows should be used

For a small project, maintaining separate configuration files may be manageable. For a company operating a large monorepo with thousands of developers, however, duplication and inconsistencies can become operational problems.

Lütke Wants Broader Agent Compatibility

Lütke’s criticism is not primarily about Claude Code’s coding capabilities. Instead, he wants the tool to recognize broader conventions used across AI coding agents.

His argument is that teams should not have to maintain separate versions of the same instructions merely because employees use different AI development tools.

The concern is particularly relevant as companies increasingly allow developers to choose among multiple AI assistants. Shopify itself currently documents integrations for several tools.

Shopify’s Multi-Agent Developer Strategy

Shopify’s public developer guidance currently lists support for:

AI Development Tool	Shopify Integration
Claude Code	Shopify AI Toolkit plugin
Cursor	Shopify plugin
Gemini CLI	Shopify extension
VS Code	Shopify AI toolkit
Codex CLI	Shopify Dev MCP
Common objective	AI-assisted Shopify development

Shopify’s own website currently instructs developers how to use Claude Code, Cursor, Gemini CLI, VS Code and Codex CLI with Shopify development tools.

That makes Lütke’s comments especially significant: the company is not simply choosing one AI coding platform. It is building an environment in which multiple agents can work with Shopify’s software ecosystem.

Anthropic Responds To Lütke

Anthropic has acknowledged Lütke’s criticism.

Thariq, an Anthropic employee involved with Claude Code, said the company is working on making Claude Code more “hackable,” including easier support for AGENTS.md and other system-prompt modifications. He also explained that Anthropic had intentionally structured Claude Code around its own system prompts and model-specific behavior.

Anthropic’s position is that model families are not necessarily interchangeable and that system prompts can have a meaningful effect on model performance.

The company nevertheless indicated that it understands the maintenance burden for teams that want a common instruction system across multiple AI tools.

Anthropic’s Proposed Direction

Issue	Anthropic Response
AGENTS.md support	Working toward easier support
Custom system prompts	More hackability planned
Different model behavior	Anthropic says system prompts can affect performance
Immediate workaround	AGENTS.md can be referenced from CLAUDE.md
Long-term goal	Greater customization of Claude Code

Anthropic said developers can currently reference AGENTS.md from CLAUDE.md as a workaround while broader support is being developed.

The Monorepo Problem Is Bigger Than One File

Shopify’s situation illustrates a broader challenge created by AI coding agents.

In traditional software development, coding standards are generally communicated through documentation, configuration files, linting systems, testing frameworks and human review. AI agents add another layer because they need instructions that shape how they interpret and modify the repository.

When several agents use different instruction formats, organizations can end up maintaining multiple sources of truth.

How “Split Brain” Can Happen

A simplified workflow could look like this:

                 SAME CODEBASE
                       │
          ┌────────────┴────────────┐
          │                         │
     Developer A               Developer B
          │                         │
     Claude Code                Other Agent
          │                         │
     CLAUDE.md                  AGENTS.md
          │                         │
          └────────────┬────────────┘
                       │
              DIFFERENT GUIDANCE
                       │
               Inconsistent Output

The risk is not necessarily that either tool produces incorrect code. Instead, two otherwise capable agents can make different decisions because they are operating with different project instructions.

For a company with thousands of developers, resolving those differences manually can become expensive.

AI Coding Agents Are Becoming Enterprise Infrastructure

The disagreement comes as AI coding agents move from experimental developer tools toward enterprise infrastructure.

Companies increasingly expect AI agents to work inside production repositories, understand internal standards and execute parts of software-development workflows.

That creates new requirements around governance.

Organizations may need common policies for:

• AI-generated code
• Repository instructions
• Security checks
• Testing requirements
• Tool permissions
• Agent access to internal systems
• Code-review workflows
• Documentation standards

The more agents a company supports, the more important interoperability becomes.

Shopify’s Position Reflects A Larger Industry Debate

Lütke’s comments also touch on a broader debate about whether AI coding tools should create proprietary layers around their models or adopt common conventions.

Anthropic has argued that model-specific instructions can improve performance because different models may respond differently to system prompts and skills. Shopify’s concern is that this model-specific optimization can create additional work for engineering organizations that use several AI systems.

The tension is therefore between tool-specific optimization and cross-tool standardization.

Two Competing Priorities

Approach	Potential Advantage	Potential Drawback
Tool-specific files	Optimized for individual AI models	More maintenance
Common instruction format	Easier multi-agent workflows	May limit model-specific optimization
Multiple formats	Maximum flexibility	Risk of inconsistent guidance
Automated synchronization	Keeps files aligned	Adds engineering complexity

The outcome could influence how enterprise software teams structure AI-agent instructions in the future.

Shopify’s AI Strategy Extends Beyond Claude

The episode should also be viewed in the context of Shopify’s broader push into AI-assisted development.

Shopify is actively positioning AI as part of its developer ecosystem, including tools that help developers build applications and customize stores. Its current documentation explicitly includes Claude Code as well as competing AI coding environments.

That means the company has a direct interest in making these tools interoperable. If developers can move between AI agents without maintaining separate project instructions, Shopify could make its development ecosystem easier to manage at scale.

The issue is therefore less about whether Shopify supports AI coding and more about how those AI tools should interact with enterprise development practices.

The Bigger Picture

Tobi Lütke’s threat to potentially ban Claude Code at Shopify illustrates an emerging enterprise problem: AI coding agents need to fit into existing engineering systems rather than creating parallel sets of rules.

The dispute also highlights the growing importance of interoperability. As developers use Claude Code, Codex, Cursor, Gemini and other agents on the same projects, companies will increasingly need consistent ways to communicate repository-level instructions. Anthropic’s willingness to explore AGENTS.md support suggests that the industry may gradually move toward more flexible and standardized agent configuration.

Looking Ahead

For Anthropic, the immediate priority will be determining how much flexibility Claude Code can provide without undermining the model-specific instructions that the company believes contribute to performance. Supporting AGENTS.md and related conventions could reduce friction for large enterprises while preserving Claude Code’s existing architecture.

For Shopify and other large software organizations, the broader question is how to govern an increasingly diverse collection of AI coding agents. If common standards emerge, developers could switch between tools with fewer workflow disruptions. If every agent continues to maintain its own instruction system, companies may have to invest more heavily in automation to keep multiple sources of guidance synchronized.