Anthropic adds support for the AGENTS.md instructions spec to Claude Code; OpenAI contributed AGENTS.md to the Agentic AI Foundation last year
Devs cheer relief from compatibility headache — Anthropic on Friday surprised the developer community by supporting rival OpenAI's mechanism for passing marching orders to AI agents.
Context & Ripple Effects
Anthropic has been building Claude Code’s agent workflow in layers, from loadable Skills instruction folders in 2025 to parallel agent teams and a managed-agent harness in 2026. AGENTS.md support makes the repository-level instruction layer less tied to Claude’s own file convention.
OpenAI’s contribution of AGENTS.md to the Agentic AI Foundation gives the format a governance home beyond either vendor. Public reaction framed Anthropic’s move as interoperability progress, while the implementation retains CLAUDE.md as the first choice when both are present.
First-order effects
- Claude Code users can reuse AGENTS.md instructions in repositories that also serve other coding agents, reducing duplicate instruction files when no CLAUDE.md is present.
- Anthropic preserves its native CLAUDE.md path while adding AGENTS.md as a configurable fallback, so existing Claude Code projects do not need to change their instructions.
Second-order effects
- Teams evaluating Claude Code and OpenAI-linked coding agents can make tool choices with less risk that repository guidance must be rewritten for each agent.
- The Agentic AI Foundation’s specification gains practical validation when a major competing agent product implements it, strengthening the case for shared repository conventions.
Third-order effects
- If leading coding-agent vendors converge on portable instruction files, differentiation shifts away from basic repository context and toward agent execution, orchestration, and controls.
- A shared instruction layer also makes instruction-file precedence and authorization boundaries more consequential, because the same repository guidance can travel across multiple agents.
The trend: Coding-agent platforms are converging on portable repository instructions while competing on how agents execute and govern work around that shared context layer.