The world is being quietly rearranged by people who write very long documents.


The title they went with ANX: Protocol-First Design for AI Agent Interaction with a Supporting 3EX Decoupled Architecture Noisy translates that to

AI agents get a protocol that cuts token use in half and keeps sensitive data away from the model


Researchers designed ANX, a new standard for how AI agents interact with computer systems, that reduces the amount of data the AI has to process by 47–66 percent compared to existing methods. In practice, this means AI agents can do the same work faster and cheaper, while keeping passwords and sensitive information out of the AI's view entirely.
Right now, when you give an AI agent a task like filling out a form, it has to see the entire screen, understand the visual layout, and process a lot of unnecessary information, which wastes computational work and creates security risks. ANX treats agent-computer interaction as a protocol problem, not a vision problem, so the agent receives only structured instructions instead of screenshots. This shifts the bottleneck from AI capability to system design, which means the same AI model can handle more work with lower infrastructure costs, and it naturally prevents the AI from accidentally leaking sensitive data because the data never reaches the model at all.
Whether this protocol gets adopted by major AI platforms and API providers over the next 12 months, or whether existing tool standards (like MCP) absorb the efficiency gains and render ANX redundant.

If you insist
Read the original →