If you have ever maintained two separate instruction files for the same repository — one for Claude Code, one for every other AI coding tool — Friday's news from Anthropic probably made your week. After 13 months of developers asking, Claude Code now supports AGENTS.md, the tool-agnostic instruction standard that OpenAI originally championed. It is a small filename change on the surface, but it says a lot about where agentic AI tooling is heading.
01.What Anthropic actually shipped
On September 18, Claude Code engineer Thariq Shihipar announced on X: "Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config." That is the whole feature — a fallback. If your project already has a CLAUDE.md, nothing changes; Claude Code keeps reading that file first, and it will not merge the two.
The implementation is actually broader than a simple filename alias. Anthropic built the support as a native "mod" — its upcoming mechanism for customising the Claude Code harness — and shipped the mod's source openly. The mod offers four instruction modes: load only CLAUDE.md, use AGENTS.md as a fallback (the default), load both formats, or restrict context to centrally managed instructions. Admins can set the policy through user settings, a --settings flag or managed settings, but notably not through a repository's own .claude/settings.json — so code checked into a project cannot dictate an individual developer's instruction policy.
02.Why developers waited 13 months for AGENTS.md
Quick background for anyone new to agentic coding: these markdown files get read by AI coding agents with every request. They spell out expected behaviour, coding conventions, build commands and testing rules — effectively the "marching orders" for the agent. CLAUDE.md is Anthropic-specific, while AGENTS.md is meant to be tool-agnostic and is already used by OpenAI's Codex, Cursor, Gemini CLI, Amp, Windsurf and Cline.
OpenAI contributed AGENTS.md to the Agentic AI Foundation under the Linux Foundation, and by December 2025 more than 60,000 open source projects had adopted it. So a team running Claude Code alongside any of those tools had to maintain two sets of instructions — or fiddle with symlinks to keep them in sync. GitHub issue #6235, asking for AGENTS.md support, was opened on August 21, 2025 and collected 6,653 reactions and 400 comments, making it by far the most-reacted request in the Claude Code repository.
03.The messy part: the issue was closed before it shipped
Here's the detail that had the community grumbling. A month before anything actually shipped, Claude Code maintainer Boris Cherny closed issue #6235 as "completed" — on August 17, 2026 — with no AGENTS.md support in the product. Users noticed immediately. "Issue is not resolved. It must be reopened," one commenter wrote on September 8, while another pointed out it was "wild that this is the most upvoted Claude Code issue by over 2x." The real fix landed nine days later, on September 18, though the issue itself was never formally reopened.
Honestly, what surprised me most wasn't the delay — it's that Anthropic caved at all. Claude Code is popular enough that the company could have kept pushing its own format. Instead, we got a genuine détente: OpenAI's head of core product, Thibault Sottiaux, responded to the announcement with "Yay! This is the way. Come to the light," and Shihipar replied with a handshake emoji. In a year full of AI labs suing, sniping and racing each other, two rivals agreeing on a shared standard feels almost heartwarming.
04.What's still missing
The fallback is not available on Amazon Bedrock, Google Vertex AI or Microsoft Foundry deployments yet, according to Anthropic's release notes — which matters, because a meaningful share of enterprise Claude Code seats run through those platforms. Teams standardised on AGENTS.md there still need a CLAUDE.md. The same release also retired the TaskOutput tool and strips invisible Unicode characters from prompts, two smaller housekeeping changes worth knowing about if you have automation built on top.
05.The bigger story: one instruction file for every agent
Zoom out and this is part of a pattern worth watching: AI coding tools are converging on shared plumbing. We covered how Zed's Delta agents started replacing pull requests earlier this week, and Claude Code's multi-agent projects before that — the common thread is that the winner won't be the agent with the flashiest demos, but the one that plays nicest with the rest of your stack. A single AGENTS.md briefing Codex, Cursor, Gemini CLI and now Claude Code means less duplicated configuration and fewer "works with one agent, breaks with another" headaches — exactly the kind of friction Indian IT services teams juggling multiple client environments know too well.
If you're just getting into agentic AI and want to go beyond config files — actually building and supervising agents that do real work — our GPT-6 Astra — Zero to Hero Guide has a full chapter on AI agents, plus sections on prompting and cost optimisation that pair well with any coding agent you run. Context files like AGENTS.md are exactly where most developers quietly lose hours — our AI Coding Playbook walks through setting them up properly, with 25 copy-paste prompts included.
More on this if you're interested: Claude Code Projects: Run a Team of AI Agents in the Cloud and Claude Fable 5.1 Is Here: Cheaper AI Agents, Fewer False Alarms, and a Science-First Twin.
Key Takeaways
- Claude Code 2.1.277 reads AGENTS.md when CLAUDE.md is absent; toggle it under "Project instructions" in /config
- CLAUDE.md keeps priority — it's a substitution, not a merge
- The feature request had 6,653 reactions and waited 13 months, and was briefly marked "completed" before shipping
- Bedrock, Vertex AI and Foundry users don't get the fallback in this release
- AGENTS.md is now read by essentially every major AI coding agent, making it the de facto shared standard
Frequently Asked Questions
What exactly changed in Claude Code version 2.1.277?
Claude Code now checks for AGENTS.md if a project has no CLAUDE.md file. It is a fallback, not a merge — when both files exist, CLAUDE.md still wins. The behavior can be toggled in /config, and the support ships as a native 'mod' whose source Anthropic released openly, with four instruction modes including both-formats and centrally managed options.
Why was AGENTS.md support such a big deal for developers?
GitHub issue #6235, opened on August 21, 2025, collected 6,653 reactions and 400 comments — by far the most-reacted request in the Claude Code repository. AGENTS.md is tool-agnostic and already used by OpenAI's Codex, Cursor, Gemini CLI, Amp, Windsurf and Cline, so teams running Claude Code alongside those tools had to maintain two separate instruction files.
Who does not get AGENTS.md support yet?
Teams deploying Claude Code through Amazon Bedrock, Google Vertex AI or Microsoft Foundry don't get the fallback in this release. Admins elsewhere can set the instruction policy through user settings or managed settings — but not through a repository's own .claude/settings.json, so code checked into a project cannot dictate a developer's instruction policy.
What This Means for You
If you're a developer in India using Claude Code at work or for side projects, update to 2.1.277, delete that symlink hack, and let one AGENTS.md brief all your agents. If you lead a team, this is a good week to standardise your repo instructions on AGENTS.md and stop maintaining parallel files. And if you want to go beyond coding agents and wire up automations for your whole workflow — n8n, Make, WhatsApp included — our AI Agents & Automation Playbook has 10 ready-to-build blueprints with verified 2026 pricing in rupees. If you're still early in your AI journey — a student or a professional moving into agentic tooling — the practical next step is learning to direct these systems properly, and the GPT-6 Astra — Zero to Hero Guide is built exactly for that, from prompting fundamentals to building your first agents.
By the way, if terms like agents and computer use still feel fuzzy, our GPT-6 Astra guide for beginners explains them all without the jargon.