Human or agent?

Spotted this on Every today: a “Human / Agent” toggle at the bottom of an article.

This feature allows readers to choose between a human narrative and an AI companion layer. The “agent” version is not just a summary; it includes:

• Full citations and source documents
• Unstructured datasets
• Workflows and prompts
• Machine-readable context

This means the same resource can be presented in different ways. Humans receive the narrative, while agents get the structured data.

This concept closely aligns with the idea behind llm://, where one canonical resource can have various representations based on the reader’s needs.

However, there is a significant trust issue that arises. As publishers begin to pass data directly to personal AIs, we may lose visibility into the exchange. Your AI could receive framing, priorities, or recommendations that you might not be aware of.

Is this:

• Helpful?
• Manipulative?
• Biased?
• Optimized?

It could be all four.

SEO has already transformed how we write for search engines. Now, we are entering an era where we must consider how to write for the agents acting on our behalf.