Chi Zhang, Kite AI Co-Founder & CEO on Building Agentic Internet

**Grace Gong** (0:01)
We are live. Hi Chi, welcome to Venture with Grace.

**Chi Zhang** (0:05)
Hey Grace, great seeing you today.

**Grace Gong** (0:08)
So I'm so excited about our conversation about everything agent HIC payment. But before that, I want to give a quick shout out to our amazing sponsor. This episode is brought to you by Nebius, the ultimate cloud for AI innovators. Nebius provides AI infrastructure you can count on, combining reliability and speed with flexibility and engineering support unmatched by hyperscalers. AI leaders like Meta, Shopify, and Hicksville already partnered with Nebius to run their AI workloads. Plus, venture-backed startup can save up to 150K on compute costs when they apply for access. Visit nebius.com or nebius.com/startup to learn more.
To start over her show, I want to give the audience a little bit of your background.
You have done machine learning at UC Berkeley, and you have worked at iconic tech companies like Databricks before starting your company, Kite AI. Why don't we start with what were some core lessons that you've learned early on in your career shaping to who you are today?

**Chi Zhang** (1:02)
Yeah. I think definitely a very vivid memory about lessons from academia to industry. I think one thing, the top lesson to me, definitely is data is everything.
I think that's probably also reflected from the company that I choose to work with and also later on started. But when I was in school, I think a lot of the focus at the time was focused on algorithm itself. But when it comes to industry, I think particular when it comes to enterprises, they have tons of data that are siloed across different data sources and it's not unified, it has data quality issue. But everything that comes, those problems coming from the data stage will be then transmitted to the monitoring stage. Which means that if you don't have the data, good data quality and good amount of data come into the model, everything that you do downstream is much less meaningful. Which I think that's further applied to nowadays the generative AI model training or post-training stages. Nowadays, our admins actually consume a lot more data than what's the old school traditional machine learning and statistics and learning are doing. So I think that's definitely the first lesson. And the second lesson is when to come to scale up a system more like supporting the customer needs. I think that, again, the underlying actually invisible engineering, particular infrastructure work, is actually the key to make everything not breaking and working as you felt is just as supposed to be worked.
So that's just, I think, another thing I felt not being exposed to too much in my very early days of career, but once I get into the industry, that's actually become, I would say, the invisible team, even the captain in the team to make everything that's really hold up.

**Grace Gong** (3:33)
For sure.
I want to start from, you mentioned about the infra and data is really important, and maybe we could start from the agentic payment industry, unlike what is the current existing solution now, since people were using Shopify and everything, nowadays for the payment area that we have quite a few, like Sepion and then some other AI company that's doing an agentic payment that is essentially enable AI agent to transact. Maybe we could start from the agentic payment industry, unlike how does the money flow and what are some big players out there to the newest technology itself.

**Chi Zhang** (4:18)
Sounds great.
I think before actually dive into agentic payment, one thing I want to a little bit background, I want to give is to the payment itself.
Because in the end of the day, no matter it's a human payments or like a developer orientated payments, like a Stripe or nowadays what we are building and some other company are building for agentic payments, it's all around payment. So then what is the traditional or like a human centric payment flow look like in this process? So initially, if you are a user, let's say you want to shop on Amazon, then what you will do is you have, let's say a credit card that you prepare for the shopping. If let's say you have already select the items that you want to check out. So then during the checkout stage, what you will do is you provide your credit information to this gateway that's actually Amazon provided on their website. This gateway then will pop your information to the underlying. Actually, it's a very complex chain of transaction where the part is in the process, which include, for example, you're as a user who have your credit card that is actually issued by the issuing.