The Agentic Automation Canvas: a structured framework for agentic AI project design
This paper introduces the Agentic Automation Canvas (AAC), a structured, machine-interoperable framework and open-source web tool designed to standardize the prospective design, governance, and evaluation of agentic AI systems by capturing six key dimensions and exporting FAIR-compliant project contracts.
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 hiring a team of super-intelligent, autonomous robots to do your job. Maybe they are scientists who write research papers, doctors who diagnose patients, or analysts who find new medicines. These aren't just simple tools; they are "agents" that can think, plan, and act on their own.
But here's the problem: How do you hire them without a contract?
Right now, when people build these AI agents, they often just start coding. They guess what the AI should do, hope it works, and then get surprised when it fails or does something weird. It's like building a house without a blueprint, or hiring a chef without telling them what ingredients you have or what you want to eat.
This paper introduces a solution called the Agentic Automation Canvas (AAC). Think of it as a "Project Blueprint" or a "Digital Contract" specifically designed for these smart AI agents.
Here is how it works, broken down into simple analogies:
1. The "Two-Sided" Conversation
Usually, the person who wants the AI (the User) and the person who builds the AI (the Developer) speak different languages.
- The User says: "I want this to save me 10 hours a week and be super accurate."
- The Developer says: "That sounds cool, but the data is messy, and the AI might get confused."
The AAC forces them to sit down at the same table and fill out a single-page form together. It's like a dating profile for a project. The User writes down exactly what they hope to gain (time, money, quality), and the Developer writes down exactly what is technically possible. If the User wants a Ferrari but the Developer only has a bicycle, the Canvas spots that mismatch before anyone spends money building it.
2. The Six "Rooms" of the Canvas
The form is divided into six sections, like rooms in a house, to make sure nothing is forgotten:
- The Definition: What is this project? (The address of the house).
- User Expectations: What do we want to achieve? (The dream renovation).
- Developer Feasibility: Can we actually build it? (The structural engineer's report).
- Governance: Who is in charge, and what are the rules? (The building codes and safety inspectors).
- Data: What information will the AI use? (The raw materials).
- Outcomes: How will we know if it worked? (The final inspection).
3. The "Magic" Feature: It's Private but Portable
One of the coolest parts of this tool is how it handles privacy.
- The "Local-First" Rule: The tool runs entirely in your web browser. It's like writing a letter on your own computer and never sending it to a server. No one else sees your data unless you choose to share it. This is crucial for hospitals or companies with secret research.
- The "RO-Crate" Suitcase: When you are done filling out the form, the tool packs everything into a digital suitcase called an RO-Crate. This suitcase is "FAIR" (Findable, Accessible, Interoperable, Reusable). It's like a standardized shipping container that any computer in the world can read. You can keep it in your safe, send it to a partner, or publish it for the whole world to see.
4. The "Living Document"
Unlike a paper contract that you sign and forget, the AAC is a living document.
- As the project grows, you update the Canvas.
- It keeps a history of changes (like a version control system), so you can see how the plan evolved from "Idea" to "Prototype" to "Final Product."
- It even includes a special file (
AGENTS.md) that acts as a instruction manual for other AI agents. If you want a coding robot to help build the project, it can read this file and understand exactly what to do.
5. The "Bridge" to the Future
The paper also mentions a future upgrade where this Canvas connects to Policy Cards.
- Think of the Canvas as the Plan (what we intend to do).
- Think of the Policy Card as the Traffic Cop (what the AI is allowed to do while running).
- The Canvas automatically translates the plan into rules for the Traffic Cop, ensuring the AI doesn't go rogue once it's turned on.
Why Does This Matter?
Right now, AI projects often fail because expectations don't match reality. The Agentic Automation Canvas is a tool to stop that. It turns vague hopes into clear, measurable promises and technical realities.
In short: It's a shared notebook that helps humans and AI builders agree on the plan, spot the risks early, and create a digital record that travels with the project from the first idea all the way to the final launch. It turns "Let's just try it and see" into "Here is exactly what we will build, how much it will cost, and how we will measure success."
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.