If you have checked your store's root URLs recently and found files you never created, that is Shopify working on your behalf. The two files answer different questions.

In early May 2026, Shopify auto-deployed llms.txt across roughly seven million storefronts. On May 28, it shipped a follow-up change that made agents.md the canonical AI-discovery file. llms.txt and llms-full.txt now default to mirrors of it. The reason for that switch explains the difference better than any format spec.

llms.txt is the older idea. Proposed by Jeremy Howard in late 2024, it is a lightweight Markdown file at your domain root. It summarizes what the site is and points language models to clean versions of key content. Think of it as a brand bio written for machines. Its companion, llms-full.txt, carries longer detail such as fuller product and inventory data.

The proposal spread widely through 2025, especially on documentation sites. Adoption by the AI companies themselves lagged. An analysis of 500 million LLM bot events found GPTBot, ClaudeBot, PerplexityBot and Google-Extended overwhelmingly skip the file and crawl HTML directly. Surveys of the standards landscape describe llms.txt as widely adopted by publishers and largely ignored by the providers it targets.

Google settled its side of the argument in its May 2026 AI guidance. It states plainly that llms.txt is not needed for AI Overviews, AI Mode or any generative Search feature.

agents.md is the newer idea, aimed at a different visitor. Emerging in late 2025, it is less a brand summary and more an operational contract. It describes what the site can do, which agent interfaces it supports (MCP endpoints, structured data feeds, supported actions), and the metadata an autonomous agent needs to interact with the store rather than merely read about it.

The shorthand that fits: llms.txt tells a model what your store is. agents.md tells an agent what it can do there.

That distinction matters on Shopify specifically. The platform's whole agentic bet, from Agentic Storefronts to Catalog syndication and AI-channel checkout, assumes agents that act: comparing live inventory, quoting return policies, completing purchases. A capabilities file serves that future. A summary file described the previous one.

Shopify's fallback chain makes the platform's opinion easy to read. agents.md is primary. The llms files mirror it unless you override them, and all three are now independently customizable through Liquid theme templates. By turning the older format into a pointer at the newer one, Shopify is signaling where it thinks agentic discovery consolidates, while staying compatible with anything still fetching llms.txt.

What should a merchant do with this? Mostly, know that the layer exists and keep it in proportion. Your actual visibility in AI shopping runs through Catalog and structured product data, which agents demonstrably consume. The root files are the address plaque on the building.

Whether to touch that plaque, customizing versus leaving Shopify's defaults, is its own question with a genuinely two-sided answer. The companion piece on customizing the defaults covers it.

The insight here stands on its own. The industry's discovery standard is migrating from files that describe websites to files that describe capabilities, because the visitor it is written for stopped reading and started doing.