Presenting The Agentic Profile at Fetch.ai Innovation Lab

I was fortunate to be invited by Steven Echtman to pitch at the Agentic AI Pitch and Demo night in SF on March 6th. This event was a great forcing function for me to take the Agentic Profile test code I'd written for Steve's February 15th hackathon, and turn that code into a proper open source library on NPMJS.

I've certainly got the "Agentic AI" bug, and am convinced that we will all have many AI agents working for each of us.

Some of our agents will interface with other agents that are designed to work together - I'll call these secondary agents subservient. Like employees of a company they do their best to serve the goals of the primary agent. This is the primary use-case today.

In the near future, some of our agents will communicate with agents of other people or companies and try to accomplish tasks together in a peer-to-peer way. Each of these agents may have conflicting goals, and the peer-to-peer relationship can be adversarial. For example, my agent may be trying to buy your car from your agent. My agent's goal is to pay the smallest amount and ensure the car is the right one for me, while your agent is trying to extract the highest price, convince me the car is right for me (no matter what) and avoid disclosing information that may cause me to shy away from the purchase.

In the age of generative AI these peer-to-peer adversarial interactions become even more interesting when each person can choose the AI models. You may be able to afford a more capable (and expensive) AI model that is better at selling me a car, and I may be on a freemium plan that will more readily agree to a higher price.

This imbalance in inferencing ability will play out across all agentic interactions: commerce, marketing and spam avoidance, business and social networking, dating, conflict-resolution, politics, health care, longevity...

What's Next For the Agentic Profile?

While this first pass on the Agentic Profile framework was clean sheet, I'm taking a step back to review the other service discover systems, authentication frameworks, and communications protocols to make sure I'm not reinventing the wheel.

Peer-to-peer (P2P) Agentic AI has unique requirements as summarized by Richard Dulude in his article The Authentication Layer for Agentic AI.

I'm really excited to be exploring P2P Agentic AI and looking forward to the day we can all have agents that easily interact with each-other to make our lives better!