You saw the changelog. Claude Code now reads AGENTS.md if there’s no Claude.md. The comments section is full of developers breathing a sigh of relief. Someone even wrote, “Finally doing something standards-compliant instead of forcing users into a proprietary workflow.”
But we need to stop patting Anthropic on the back. Standards aren’t charity; they’re a land grab.
Anthropic trading a proprietary file convention for an emerging ecosystem standard isn’t a surrender of control. It’s a pivot. By adopting AGENTS.md, they aren’t giving up their dominance; they are embedding themselves deeper into your workflow. The war for AI isn’t about who has the best model anymore. It’s about who owns the context layer.
Look closely at what actually changed. The update is intentionally partial. Claude Code still ignores .agents/skills. They are maintaining a tight grip on the agent workflow stack while letting you think you’ve won a battle for openness. It’s a brilliant illusion.
When you control the context layer, you don’t need to own the code—you just need to own the interpreter.
If you build or maintain agent workflows, you’re probably groaning about having to redo your scaffold harness. You should be groaning. But not just because of the refactoring. You should be groaning because you’re being funneled into a new type of lock-in. You’re standardizing your repositories around whoever executes the convention best. And right now, that’s Anthropic.
Most people see this as Anthropic ‘doing the right thing.’ The sharper read is that AGENTS.md becomes a moat. They let the open-source community build the road, and now they are setting up the toll booth.
Don’t mistake a strategic retreat for surrender. The company that defines how an open standard is interpreted doesn’t just join the ecosystem—they become the ecosystem.
FAQ
Q: Isn't adopting AGENTS.md a step toward openness?
A: It's a step toward interoperability, but not out of the goodness of their hearts. Anthropic is adopting the standard because they realize format control is dead; owning the interpreter of the context layer is the new battleground.
Q: What does this mean for my current agent workflows?
A: You'll have to rework your scaffolding to support AGENTS.md. More importantly, you have to decide whether to trust this emerging convention or keep your proprietary harness, knowing the ecosystem is standardizing around Anthropic's interpretation.
Q: Is AGENTS.md just another form of vendor lock-in?
A: Yes. The standard might be open, but the execution is proprietary. By selectively ignoring parts like .agents/skills, Anthropic ensures they still control the agent workflow stack, making the standard itself a competitive asset.