Anthropic Adopts OpenAI's AGENTS.md Standard, Easing AI Agent Instructions for Developers

Anthropic Embraces OpenAI's AGENTS.md Standard in Claude Code

Anthropic surprised the developer community by announcing support for OpenAI's AGENTS.md format in Claude Code. Starting with version 2.1.277, Claude Code will check for and use AGENTS.md if no CLAUDE.md file exists in a folder, according to Claude Code engineer Thariq Shihipar1

1

. This decision marks a significant shift toward interoperability between AI platforms and reduces the maintenance burden developers face when working with multiple AI coding agents.

The move addresses a longstanding pain point for developers who use Claude Code alongside OpenAI Codex or other non-Anthropic tools. Until now, these developers had to maintain two separate sets of markdown instructions for their AI agents: CLAUDE.md for Anthropic's tools and AGENTS.md for everything else2

2

. These instruction files outline expected behavior, preferred tool usage, coding conventions, and establish the foundation of the agent's context with every request.

Reducing Developer Maintenance and Operational Overhead

The support for AGENTS.md eliminates redundant work that plagued development teams. Developers previously relied on workarounds like creating symlinks to keep CLAUDE.md and AGENTS.md synchronized, but these solutions became increasingly problematic in larger codebases2

2

. Amit Kumar Jena, AI development head at IT consulting firm Kanerika, noted that this change could "remove some real busywork" for development teams working with multiple coding tools2

2

.

In monorepos with instructions at multiple directory levels, maintaining separate files created substantial developer maintenance debt that often went unnoticed under deadline pressure until tools quietly drifted out of sync2

2

. By adopting a common instruction format, teams can now use AGENTS.md across their coding tools without duplicating effort or risking inconsistencies between instruction sets.

Standardization in AI Agent Instructions Gains Momentum

OpenAI contributed AGENTS.md to the Agentic AI Foundation under the Linux Foundation in a bid to build support for its standard. As of December 2025, more than 60,000 open-source projects had implemented AGENTS.md1

1

. The popularity of Claude Code had given Anthropic the latitude to insist on its own standards, but the company appears to have concluded that continued specification separatism offered no strategic advantage1

1

.

Thibault Sottiaux, OpenAI's head of core product, welcomed Anthropic's decision with enthusiasm, remarking "Yay! This is the way. Come to the light," to which Shihipar responded with a handshake emoji1

1

. This public exchange signals a collaborative approach to standardization in AI agent instructions that benefits the broader developer ecosystem.

Enterprise Governance and Compliance Benefits

For enterprises, a common instruction format creates practical advantages beyond reduced operational overhead. Charlie Dai, principal correspondent at Forrester, explained that standardization "can become a practical control point for coding conventions, security requirements, architectural guardrails, and governance policies"2

2

. This centralized approach to AI agent instructions makes it easier for organizations to enforce consistent standards across development teams while maintaining compliance requirements.

Deployment Limitations and Configuration Options

Despite the progress toward interoperability, some deployment-specific limitations remain. AGENTS.md support depends on Claude Code fetching feature flags from Anthropic. If Claude Code cannot retrieve these feature flags—for instance, when deployed through Amazon Bedrock, Google Vertex AI, or when telemetry is disabled—the tool falls back to CLAUDE.md2

2

. This limitation affects enterprises using third-party providers even though the instruction file itself is local and requires no network connection to be read.

However, Claude Code provides flexibility through configuration options. Users can toggle AGENTS.md behavior using the /config command, allowing teams to customize which files Claude Code reads, including AGENTS.md, CLAUDE.md, or both1

1

2

2

. This gives development teams a way to use AGENTS.md for instructions shared across coding tools while retaining CLAUDE.md for Claude-specific features.

Compatibility Considerations for AI Coding Agents

While the common instruction format streamlines workflows, developers should note that CLAUDE.md includes Claude-specific features that AGENTS.md was not designed to support. Instructions written for one agent, such as Codex, may not behave identically when Claude Code reads them2

2

. This means AI coding agents aren't fully interchangeable despite sharing a common instruction format, and teams may need to account for platform-specific behaviors when crafting their agent instructions.