Anthropic's Claude Code Faces Backlash Over Refusal to Adopt AGENTS.md Standard

A dispute has emerged over the interoperability of AI coding tools after Anthropic declined to support AGENTS.md, an emerging open standard for project-level AI context, in its Claude Code tool. The conflict centers on whether AI agents should rely on unified standards or proprietary configuration files.

The friction came to a head on the Claude Code GitHub repository, where user DylanLIiii opened a feature request (Issue #6235) urging Anthropic to adopt the AGENTS.md format. Currently, Claude Code utilizes a project-specific file called CLAUDE.md to provide instructions and context to the agent. DylanLIiii argued that this approach is too narrow, stating that CLAUDE.md "doesn’t work as well when collaborating with other developers who aren’t using Claude Code."

The Push for Standardization

AGENTS.md is designed to function as a "README for AI agents," providing a single, unified source of codebase understanding that can be read by various AI tools. As the market for AI coding assistants expands to include tools like Cursor and Codex, the industry is seeing a growing push toward standardized context files. The goal is to prevent a fragmented environment where developers must maintain multiple, redundant instruction files—such as .cursorrules or CLAUDE.md—to accommodate every different AI tool used across a development team.

The Cost of Proprietary Ecosystems

This standoff highlights a broader tension between open standards and the "walled garden" approach to AI development. When major providers like Anthropic opt for proprietary formats over cross-tool standards, it increases operational friction for developers who employ multi-model workflows. By requiring a tool-specific file, the ecosystem risks creating vendor lock-in, where the cost of switching tools or collaborating across different platforms becomes prohibitively high due to the need to rewrite project context.

Future Outlook

While some developers suggest that the gap can be bridged using technical workarounds like symlinks, the core issue remains a matter of strategic alignment. It remains to be seen if community pressure or the widespread adoption of AGENTS.md by other major AI labs will eventually force Anthropic to move away from its proprietary CLAUDE.md format. For now, the industry remains split between those favoring a unified AI-readable layer and those building tool-specific optimizations.