Dialogue with Kite AI: How to create a unified framework for payments, identification, and governance for AI Agents?

Interview: The Round Trip

Compiled & Organized by: Yuliya, PANews

As the wave of AI sweeps across the globe, a new era of the internet dominated by machines (such as AI Agents) rather than humans is quietly emerging. To enable machines to collaborate smoothly and securely in this new world, a unified identity, payment, and governance programmable trust infrastructure has become crucial. Kite AI has emerged in this context, dedicated to building the world's first AI-driven payment blockchain network, and has already attracted the attention of top investment institutions such as Paypal Ventures, Coinbase Ventures, and General Catalyst.

In the new series Founder’s Talk of The Round Trip co-produced by PANews and Web3.com Ventures, host John Scianna and Cassidy Huang invited Chi Zhang, co-founder and CEO of Kite AI. She shared her journey from top AI and data companies to entrepreneurship, Kite AI's grand vision of building a trust framework for the machine internet, and how to seize the “once-in-a-lifetime” opportunity in a new AI paradigm composed of data, computing power, models, and agents.

Building Trust for the Machine Internet: The Vision and Mission of Kite AI ###

Host: Welcome Chi, could you start by introducing yourself and sharing your vision for Kite AI?

Chi Zhang: I have a background in AI and big data. I obtained my PhD in machine learning and artificial intelligence from the University of California, Berkeley. After graduation, I joined DotData, which was one of the earliest and largest automated machine learning platforms at that time, where I was responsible for leading data science and product-related work, serving various verticals such as healthcare and finance (for example, using AI for trading fraud detection in banking and using AI for medical image-assisted diagnosis in healthcare). Later, I joined Databricks, where I managed product solutions for data engineering. Subsequently, I co-founded Kite with co-founder Scott.

Our mission with Kite is simple: we believe that the future of the internet will be dominated by machine roles (such as AI Agents), and their number will even surpass that of humans, which is becoming a reality.

In order for these machines to seamlessly and smoothly complete various workflows on the Internet for humans, businesses, and organizations (such as helping you purchase daily necessities or assisting companies with recruitment and interviews), these AI Agents must possess three core abilities:

• Identity: Having identity verification to prove “who they are”.
• Payment: The ability to pay and receive funds instantly and securely.
• No Loss of Control: We want everything they do to be under clear limits and guidance to ensure they do not lose control.

So, this is basically the original intention and goal of founding Kite: to build a programmable trust infrastructure, or a framework that unifies identity, payment, and governance in a programmable way, allowing AI Agents to truly represent humans or any entity to perform tasks under clear guidance.

Host: Was there a moment in your previous work experience when you had a “lightbulb moment” that gave you unique insight and made you realize, “This is the direction we must move forward in”?

Chi Zhang: I used to believe that data was one of the four core pillars of AI (in recent years, when discussing the pillars of AI, the dimension of “Agent” has gradually been added to “data, computing power, and models”). During my PhD studies, I focused on training models and causal inference, and later I delved into data engineering at Databricks. I personally experienced that data is actually the biggest bottleneck for companies doing AI—especially high-quality, unique, and novel data is crucial for training models. Even though many people consider computing power as the bottleneck (this is also one of the reasons why Nvidia's stock price surged in the past two years), I still believe that data is an important and urgent issue, especially when we talk about Agents.

And now, when we talk about AI Agents, if you go and communicate with companies that focus on Agents or data infrastructure companies, they will all point out that one of the most urgent and severe bottlenecks to truly making Agents work is solving the underlying data infrastructure issues. For example, if you want an Agent to help you with cryptocurrency trading, it needs real-time access to a large amount of data, such as API data for coin prices, as well as sentiment data from various social platforms like Twitter. This requires Agents to have the ability to access and process unstructured or semi-structured data in real-time, which remains a bottleneck in the current data infrastructure.

Returning to your question about when I first saw this unique opportunity and combination, I believe it really began in the second half of last year. At that time, we started to see significant improvements in the capabilities of Agents, which became particularly clear at the beginning of this year. At that time, Manus launched a general-purpose AI Agent, OpenAI released the ChatGPT Operator, and many other companies showcased tremendous advancements in the workflow, accuracy, and intelligence of Agents, often requiring minimal human instructions.

Taking all of this into account, the inspiration for me and our team is: if we are not going to create a computing power company like GPU cloud services, nor pursue a business like Scale AI or Databricks (although I still believe they are among the greatest companies in the industry), then what is the next truly significant opportunity that we, as a startup team, should focus on?

I believe the answer lies in AI Agents or agent-based infrastructure. This can be described as a “once-in-a-lifetime” opportunity. Because if you look back at the development of the internet, it has been over 30 years since 1995, and it has primarily been built for humans—ranging from desktop internet to mobile internet. Many designs, such as authentication, the need to enter a CVV code when using a credit card, and the aesthetic design of website frontends, are all aimed at optimizing the visual, auditory, and overall experience for humans.

But I believe that the form of the internet will be completely different for future “machine actors” or AI Agents.

Kite AI's positioning in the agent payment ecosystem

Host: It seems that “agent payment” has become a focal point of interest for many people, with companies like Visa, Google, and the x402 protocol jointly released by Coinbase earlier this year making moves in this area. So, does Kite AI have a clear directional preference? How do you view payments between agents?

Chi Zhang: We warmly welcome and agree with the emergence of these open standards, as they are public goods that contribute to industry development. Simply put, the system we have built is 100% compatible with these open standards (such as x402, A2A, AP2), which are very important for Agent-related transactions and payments.

Our focus is more on the underlying infrastructure: building a foundation for settlement and verification layers, allowing all payment transactions or identity verification operations related to Agents to be completed on our layer. You can understand it this way: x402, A2A, and AP2 protocols are more like different token standards such as ERC-20 and ERC-721 in the Ethereum ecosystem. And we, Kite AI, are like the Ethereum blockchain itself, which is the underlying platform built to execute Agent transactions (including payments) under these 'standards.'

We are excited about the emergence of these standards and believe that public goods driven by large companies typically need major firms to promote them, rather than relying solely on startups to do the promotion.

Strategic Financing and Market Approach: The Value Behind the $33 Million

Host: You recently completed a funding round of up to $33 million, with investors including Paypal Ventures, General Catalyst, and Coinbase Ventures, as John just mentioned. We are curious about how this funding will specifically help you accelerate the realization of your ambitious product roadmap?

Chi Zhang: If you look at our equity structure table, you will find that most of the investors we choose are for strategic cooperation considerations, many of which are corporate venture capital. Even giant financial VCs like General Catalyst are known for providing a lot of “hands-on” assistance to their portfolio companies. Therefore, almost every investor we bring in has its strategic significance behind it.

Some can provide us with distribution channels or have the potential to offer these channels; some can bring us key networking resources to help us connect with partners like Agent service providers; while others have regional coverage capabilities, such as Japan's SBI Group, which can effectively assist us in achieving growth in the Japanese market.

So what I want to say is that while capital is certainly important, the network, resources, and the value that the investors on our equity structure can bring are the more interesting and exciting parts for us.

This also brings us back to the question of how to accelerate the roadmap. I believe that for any infrastructure company to promote the adoption and application of its technology, it must start from several key verticals. You need to focus on one to three tracks where you believe applications can explode the fastest. This is also why our investor composition is so important. For example, Paypal is an outstanding representative of online commercial payments, especially in the demand for e-commerce payments. Therefore, e-commerce Agent is also one of the areas we are focusing on. We believe that in the Agent payment scenario, a huge opportunity is to assist users (whether individual consumers or businesses) in completing purchases through Agents. For individuals, this could be buying daily necessities, booking flights, and hotels; while for businesses, it is more like corporate procurement behaviors, such as purchasing office supplies, or helping an automotive company procure automotive parts and manufacturing components globally.

So, this is more about our market entry strategy. The funds obtained from investors will primarily be used to help us execute this strategy, and of course, to recruit top talent in AI and other fields in Silicon Valley and other regions.

Host: Since you have so many investors with corporate backgrounds, how will they influence your product roadmap? For example, will they directly tell you their needs? Is this one of the reasons that prompted you to decide to collaborate with Brevis? Because I understand you will use their ZK (Zero-Knowledge) technology to develop the “ZK Passport.”

Chi Zhang: We have a deep connection with Brevis. In fact, before Brevis and Kite emerged, we already knew Michael (the founder of Brevis). At that time, he was still working on Celer Network, and I was working on other projects; we were friends. My co-founder's graduate school classmate is also his classmate, so we've had a long-standing relationship.

The core of payment lies in trust, and the foundation of trust is identity. This concept is inspired by discussions with industry giants like Paypal and Visa. As the world's largest personal identity network, Paypal's success validates the importance of identity in the payment system. Kite aims to create a programmable trust infrastructure that integrates identity, payment, and governance, forming a “trust layer” to support the entire payment system. In this process, ZK technology has become an important tool for achieving identity verification and privacy protection. Brevis's solution provides critical support at the verification level, while also offering a state channel-like processing solution for high-frequency trading scenarios.

The Kite team realized at the early stages of project conception that the trading or payment needs of Agents would occur at machine speed, rather than human speed. This trading model requires ultra-high frequency, real-time, and high throughput, but current mainstream blockchain technologies, including Solana and Ethereum, cannot meet such demands. To achieve real-time and low-cost intensive high-frequency trading, the Kite team explored state channels as a solution. During this process, Brevis technology came into the team's view, as its technical characteristics highly align with Kite's core needs, providing the possibility for promoting efficient trading models.

Unique challenges in building scalable architecture

Host: It sounds like you are building a well-thought-out solution. So, do you currently have design partners helping you shape this system together? Or, are the ideas you just mentioned already a consensus in the industry?

Chi Zhang: Actually, I want to answer from two aspects. First, we do have some design partners involved, some of whom are on our investor list, while others are not, but they all bring very exciting application scenarios and collaboration opportunities that we are actively promoting.

Second, regarding whether this is “consensus” or “common sense,” I would say “yes, and no.”

The part that is “is” lies in the fact that many related technologies and concepts actually already exist.

But what I mean by saying “no” is that you need a very unique combination of perspectives, along with a deep understanding of multiple fields such as blockchain, agents, payments, etc., to truly see how to correctly combine these elements together. This is also why I believe that there aren't many people who can really construct this or design a truly suitable architecture to solve this problem.

For example, if you talk to people who work in infrastructure or architecture, they will tell you that in the early stages of a project, there may be thousands of ways to architect a system to solve the same problem. However, as the system scales up and traffic surges, those thousands of methods will quickly reduce to possibly only 10 or even 5 viable paths.

We put in a lot of thought, effort, and practical testing to build one of the ultimately successful “5 Methods” from the very beginning, rather than one of the “1000 Methods” that will collapse after the system scales.