Open source coding agents are converging on a single instruction file format called AGENTS.md, allowing developers to move between tools like Aider, Cline, Continue, and Gemini CLI without rewriting their setup. This standardization, stewarded by the Agentic AI Foundation under the Linux Foundation, represents a fundamental shift in how developers think about AI tool lock-in.

Why Are Developers Suddenly Portable Between Coding Agents?

The AGENTS.md format is not a new invention; it is described as "a simple, open format for guiding coding agents" that has already been adopted by roughly two dozen tools. This compatibility list includes Codex, Cursor, Aider, Gemini CLI, opencode, Zed, Warp, Devin, Windsurf, and GitHub's coding agent. The notable absence is Claude Code, which uses its own CLAUDE.md format instead.

Developers are deliberately preparing for this portability. One developer articulated the strategy plainly on Hacker News, stating: "One main thing is to de-couple the repos from specific agents e.g. use.mcp.json instead of 'claude plugins', use AGENTS.md (and symlink to CLAUDE.md) and so on. I love this because I have absolutely 0 loyalty to any of these companies and once Anthropic nerfs I just switch to OpenAI, then I can switch to Google and so on. Whichever works best". Another developer framed the motivation differently: "I don't want my workflow to break because a company changes its usage policies".

This shift reflects a broader anxiety in the developer community about vendor lock-in. As AI coding tools proliferate and pricing models shift, developers are taking steps to ensure their work remains portable across platforms. The AGENTS.md standard provides a technical foundation for that independence.

What Are the Real Limitations of This Portability?

The portability story has important caveats that developers should understand before committing to the standard. Two of the listed tools, Aider and Gemini CLI, only read AGENTS.md after an explicit configuration line, meaning support is not uniform across all platforms. Additionally, the common workaround of symlinking AGENTS.md to CLAUDE.md carries a measurable cost in token usage.

One developer tested this assumption rather than accepting it at face value, discovering that Claude may read the presence of an AGENTS.md file in directory listings and attempt to read it directly, doubling token impact and context bloat. This means the migration bill is not your instructions themselves, but rather your permission configuration and muscle memory, which no standard currently covers.

How to Prepare Your Coding Projects for Agent Portability

• Adopt AGENTS.md as your instruction file: Write your agent instructions in a single AGENTS.md file at your workspace root, which Meta Muse Code and roughly two dozen other tools will automatically discover and load.
• Test symlink performance before deploying: If you plan to symlink AGENTS.md to CLAUDE.md for compatibility, run local experiments to measure token impact rather than assuming the cost is negligible.
• Document your permission rules separately: Keep your configuration and permission settings in version control outside of agent-specific formats, so switching tools does not require rewriting your access controls.
• Verify explicit config requirements: Check whether your chosen tools require explicit configuration to read AGENTS.md, since Aider and Gemini CLI do not auto-discover the file without additional setup.

What Happens When Agents Read Each Other's Config Files?

The portability conversation has surfaced a privacy concern that developers should be aware of. Meta's Muse Code automatically loads personal instruction files from ~/.codex and ~/.claude directories outside the selected workspace and sends their contents in the first provider request. A file stored in ~/.claude was written for Anthropic, and a file stored in ~/.codex was written for OpenAI, yet Muse treats the existence of those files as permission to copy their contents to Meta.

Meta displays a notice and provides an opt-out, but does not ask for permission before sending the files. The distinction being drawn is between a workspace file, which you deliberately chose to put in the repository you opened, and a global config directory outside it. Notice-and-opt-out is not the same as being asked. Grok Build reads the same rival configuration files and is equally upfront about it, so this appears to be a category-wide habit worth checking rather than one vendor's isolated behavior.

The fair counter, noted by developers in the same discussion thread, is that cross-reading rule files is already routine across the industry. One developer pointed out that "Codex literally reads Claude.md, this isn't even news," suggesting that the practice of agents reading competitor configuration is becoming normalized. Treat this as a category habit worth understanding when you choose which agent to use, rather than a unique privacy violation by any single vendor.

The emergence of AGENTS.md as a de facto standard signals that the open source coding agent market is maturing beyond individual vendor ecosystems. Developers are no longer willing to accept tool lock-in as inevitable, and the tools themselves are responding by supporting a common format. Whether this standardization will hold as the market consolidates remains an open question, but for now, it represents a meaningful shift in developer power and choice.