ANX: Protocol-First Design for AI Agent Interaction with a Supporting 3EX Decoupled Architecture
This paper introduces ANX, a protocol-first framework featuring a 3EX decoupled architecture and agent-native design that significantly reduces token consumption and execution time while enhancing security and reliability for AI agent interactions compared to existing GUI automation and MCP-based methods.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to teach a very smart, but slightly clumsy, robot assistant how to do your taxes, apply for a job, or book a flight.
Right now, we have two main ways to do this, and both have big problems:
- The "Human Mimic" Approach (GUI Automation): You tell the robot, "Look at the screen, find the box that says 'Name,' click it, type your name."
- The Problem: This is like asking the robot to read a book by staring at the ink on the page. It's slow, uses a lot of energy (money), and if the font changes slightly, the robot gets confused and crashes.
- The "Pre-Installed App" Approach (MCP/Tools): You give the robot a specific tool for every single task. "Here is the 'Job Application Tool,' here is the 'Tax Tool.'"
- The Problem: The robot has to download and install these tools before it can use them. If you need a tool for a website that doesn't exist yet, the robot is stuck. Also, you have to hand the robot your password so it can use the tool, which feels risky.
Enter ANX: The "Universal Remote Control" for AI.
This paper introduces ANX (AI Native eX), a new way for humans and AI to talk to each other. Think of ANX not as a tool, but as a universal language and a new operating system designed specifically for robots, not humans.
Here is how ANX works, using simple analogies:
1. The "Blueprint" Instead of the "Painting" (ANX Markup)
Currently, when an AI looks at a website, it sees a giant, messy painting (the visual screen). ANX gives the AI a clean architectural blueprint.
- Old Way: "Find the blue rectangle near the top left that looks like a button." (Hard to describe, uses lots of words).
- ANX Way: "Execute command:
submit_formwithname: Mingze." - The Benefit: It's like sending a text message instead of mailing a 50-page letter. It uses 47% to 66% less "energy" (tokens) and is much faster because the AI doesn't have to guess what it's looking at.
2. The "Magic Marketplace" (ANXHub)
Imagine you need a specific wrench to fix a sink.
- Old Way: You have to buy a toolbox, go to a hardware store, and hope they have that exact wrench. If they don't, you can't fix the sink.
- ANX Way: You have a Magic Marketplace (ANXHub). You just say, "I need a wrench for a 2024 sink." The marketplace instantly finds the perfect wrench, hands it to you, and you use it immediately. You don't need to buy the toolbox or install anything.
- The Benefit: The AI can instantly find and use tools for any task, even ones created yesterday, without needing to download anything first.
3. The "Secure Vault" (Security)
This is the most important part.
- Old Way: To fill out a form with your credit card number, you have to tell the AI, "Here is my credit card number." The AI sees it, remembers it, and might accidentally leak it.
- ANX Way: Think of ANX as a secure vault with a glass window.
- The AI sees the form (the window) and knows what needs to be filled.
- But when you type your credit card number, you type it directly into a secure vault (the Core) that the AI cannot see.
- The AI only gets a "receipt" saying, "Payment successful." It never sees the actual number.
- Human Check: For dangerous actions, the system forces a "Human Only" button. The AI cannot click it; a real human must press it. It's like a safety switch that the robot can't bypass.
4. The "Team Captain" (3EX Architecture)
ANX splits the work into three clear teams so the AI doesn't get overwhelmed:
- The Planner (Expression): Decides what needs to be done.
- The Scout (Exchange): Runs to the Magic Marketplace to find the tools.
- The Worker (Execution): Actually does the work.
- The Benefit: In the old way, one confused robot tried to plan, find tools, and work all at once. ANX separates these jobs, making the whole process smoother, faster, and less likely to make mistakes.
The Real-World Test
The authors tested this by asking an AI to fill out a job application form.
- The "Human Mimic" (Old Way): Took 33 seconds and used a lot of computer power.
- The "Pre-Installed Tool" (MCP): Took 30 seconds and still used a lot of power.
- ANX: Took only 13 seconds and used half the computer power.
Summary
ANX is like upgrading from a robot that tries to "see" and "click" like a human (slow, expensive, risky) to a robot that speaks a special, efficient code (fast, cheap, safe). It allows robots to find tools instantly, keeps your secrets hidden in a vault, and lets humans and robots work together on complex tasks without getting confused.
It's the difference between giving a robot a magnifying glass to read a menu versus giving it a digital menu that it can read instantly and order from directly.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.