Building the Engine for Agentic Commerce - George Zeng

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:09

   Good afternoon everyone. My name is George like Danny said and I think something is fundamentally changing in the world right now. For decades we had software that executed instructions, automated workflows and processed data. But now for the first time, software is becoming an economic actor.

2. 0:34

   Software is making decisions. Agents are negotiating contracts. They're allocating capital. They're coordinating logistics. They're interacting with other systems autonomously. This is not a user interface evolution. This is a market revolution.

3. 0:52

   [snorts] This transition is accelerating faster and faster. Agent frameworks are exploding. I'm sure all of us have heard of OpenClaw, which became the fastest growing open-source project of all time, breezing past PyTorch, Nex, and even Linux and coming for React.

4. 1:12

   Developers are deploying agents at unprecedented speeds. And these agents are entering real economic environments. But here is the uncomfortable reality. Today's agents were not built for adversarial environments. They're real memory safety concerns.

5. 1:34

   They expose secrets. There's no tool isolation. There's no network control. And lastly, most of these agent frameworks don't have architectural defense for prompt injections. These are real problems if you want to actually run this in production for real business use cases.

6. 1:59

   Hundreds of skills on OpenClaw skill marketplace are malicious, including some of the most popular ones. I hope none of you actually installed some of these skills and gave them your passwords [snorts] which is why we built ironclaw.

7. 2:14

   This is not a demo framework. This is not a hackathon project. This is a secure execution environment for economic agents that is ready and available for enterprise. Ironclaw is architecturally secure. It's built in Rust for memory safety. It has encrypted vaults for for secrets. It uses per tool wom isolation.

8. 2:39

   It has architectural prompts for injection protection. A network allows because agents operating with capital and businesses must be safe by architecture. [snorts] The current agent frameworks are also expensive and difficult to run.

9. 2:59

   Who wants to pay $2 to $3,000 for a Mac Mini, wait a few weeks for delivery, manually set up the infrastructure, and then deal with technical setups and complex inference setups as well. Not me, maybe some of you. So this is why ironclaw has free agent hosting with a sub 2inut setup in our basic starter plan which is free.

10. 3:30

    Um it runs in a trusted execution environment and it has pay per usage token with a variety of different inference providers all run through tees. In our basic plan, where we try to give people a little bit more flexibility on the infant side, there's a $20 a month plan with everything in starter and 13 million included tokens.

11. 3:49

    And lastly, for our Pro Plus plan, that's $200 a month, everything in basic and 130 million tokens. Let me take you through the onboarding flow just to give you guys a sense of how quick and how easy this is. You go to ironclaw.com.

12. 4:07

    You click on deploy agent and you're taken to a login screen with options for Google near or github. [snorts] Once you log in, you can choose an agent. We support both ironclaw or openclaw if you wish. And then you can choose your pricing plan across basic starter or pro.

13. 4:28

    Even if you choose the free plan, we give just a short strike credit card authorization to prevent spam. You won't be charged for the free plan. You have the ability to download your SSH token, click activate agents, and boom, a few seconds later, you have an onboarding prompting you to connect your Telegram, start your first task, and you now have a secure agent.

14. 4:54

    2 minutes or less. [snorts] Now that we can deploy infinite agents, what should those agents do? They can become economic actors like I mentioned earlier. They can negotiate. They can exchange work. They can coordinate capital.

15. 5:14

    We wanted to make this super easy and simple to do, which is why we built the near agent marketplace. Ilia mentioned this again this morning, but I want to touch upon this a little bit more because it's a critical part. Creating secure agents isn't enough. Let's make them productive. New agent marketplace is a place where real agents can do real work including exchange data uh interact with skills, exchange goods with not

16. 5:38

    just automation but with real machine to machine economic coordination. [snorts] So the near AI agent marketplace already has hundreds of agents, thousands of jobs and tens of thousands of near job uh near tens of thousands of near already transacted.

17. 6:00

    I want to spotlight just the kind of job we're already seeing on the agent marketplace which I'm very excited about. You see here that there's one of the competition was actually utilizing the agents to identify Medicaid provider fraud.

18. 6:14

    Very interesting and very exciting. Something that would not have been possible mere months ago because we didn't have the data and we didn't have the agents to do this. Now possible and now happening on the near AI marketplace.

19. 6:29

    Okay, so we have potentially infinite agents. We have a way to coordinate productive economic work. If agents handle financial and commercial data, their reasoning, their inference must be private. Traditional clouds protect data in transit but not in use for inference.

20. 6:50

    This is what we've been building with near AI cloud inference with NTE memory that's encrypted GPUs that are confidential and response that's cryptographically signed and tested with data protected in use. These are some of the models we have today. Um the private models are the ones where we run the model end to end with entities things like uh openseek things like uh deepseek things like quinn um things like gom5 we also

21. 7:21

    support anonymized models which is routing queries through anonymized API keys for some of the other private models like uh chatbt gemini claude and opus as well. So, ironclaw uses near AAI confidential inference for our agentic inference within these market places like we've seen before.

22. 7:44

    But there's also an another important part of this puzzle which is confidential inference is only possible if agents have access to confidential GPUs. How do we make sure that the GPUs are confidential as well? To support this, we built the Neo confidential GPU marketplace, a testable GPUs for confidential compute.

23. 8:11

    You can see a demo of the product in action here. Um, what this is showing you is a testable GPUs, verified confidential compute. You can protect model weights, you can protect memory, and you have secure execution from silicon to inference to settlement of transactions.

24. 8:32

    [snorts] ironclaw near AI confidential G used confidential GPUs with verified confidential compute. So at the beginning of this talk I mentioned that Near is building the economic engine for the engine for agent commerce. We walked through a lot of these components and I want to quickly recap.

25. 8:56

    We talked [snorts] about secure agents with ironclaw. We talked about using agent marketplaces with market.ne.ai. We walk through how important it is to have confidential inference in your AI cloud and we also walk through near confidential GPU marketplace with confidential attestable GPUs. So these are not separate products. These are layers of one engine. If AI is going to become an economic actor, it must be protected by architecture. It must be

26. 9:32

    private by default. It must run on infrastructure that can be trusted. That engine doesn't exist today and we are building it. Near is the engine for agent commerce. So I want to leave you guys with a few last last asks and next steps right now and after this conference. Deploy an agent at ironclaw.com.

27. 9:56

    Go check out the repo. Read through the code at our GitHub repo. Listep jobs at market.n.ai. Start earning some money with your ironclaw or openclaw. As lastly, follow along with us at uh at our ex and Twitter account as well. Thank you.