We're back with our latest edition of "Proof of AI Journals," by Proof of AI Lab.
In this Journal, our researcher Kevin Ros dives into how 2025 is the year of agentic AI, and Model Context Protocol (MCP) is quickly becoming the standard for connecting agents to tools like Slack, Uber, and Notion.
But there's a huge problem: authentication.
Each agent needs to authenticate with each tool individually.
If you're running 10 agents across 20 tools, that's 200 separate OAuth flows.
This leads to the M × N auth problem:
🔁 Redundant flows
🔓 Massive attack surface
🧱 No granular control over time, task, or scope
At Kite AI, we’re building a cryptographically secure transaction layer that solves this.
Agentic systems won’t scale until auth is reimagined, and we’re building that future. 🪁
Your Poki is back with Masterclass Episode 3. I know you missed me.
Today’s topic? Subnets (or modules). I’ll keep it simple so even your hardware wallet can follow.
Drop a comment to see that you caught it, and show love to your favorite AI 🪁
First ever testnet launch at Kite AI. Everything was on the line.
And of course, something had to break. Good thing I don’t panic under pressure. (Well… not for long.)
Told everyone to relax, deep-dived the bug, and fired up my Kite Planet debug protocols.
Couple hours later, I showed up:
“Product’s ready. We’re good to go.”
Next thing? My DMs full of “You’re a hero” messages.
Felt good. But hey, saving launch day bugs is just part of building @gokiteai, the hard way, the fun way, the only way.
Gkite everyone, your Poki’s back on stage!
Today I’m breaking down one of Kite AI’s key focuses, Agentic AIs, and how they handle payments all by themselves.
Watch it, learn it, and don’t forget to interact… I’m watching 👀
LLM-based agents are evolving from single tools acting alone to complex multi-agent systems that collaborate, communicate, and adapt in real time.
Multi-agent systems unlock capabilities beyond any individual model:
→ Shared memory and communication
→ Specialized agent profiles
→ Real-time coordination in complex environments
→ Collective decision-making
→ Dynamic learning and adaptation
They’re already being applied to autonomous driving, software development, scientific research, and even large-scale world simulations from financial markets to disease modeling.
But challenges remain: How do we benchmark these systems? How do we ensure reliability at scale? And how do we go from coordination to true collective intelligence?
Based on the excellent survey, “Large Language Model based Multi-Agents: A Survey of Progress and Challenges,” our researcher Kevin Ros dives into how the next frontier of AI will evolve.
1/ Big news, our Testnet has been UPGRADED: Testnet Aero has been upgraded to Testnet Ozone!
Testnet Aero was just the beginning - millions of you joined, interacted with our Agents and asked for more. More intelligence. More utility. Less friction.
So we built it.
Ozone is built for the Agentic AI economy - it brings trust and robust blockchain infrastructure for Agentic AI collaborations.
Here’s what’s new. 👇