This session from MCP Dev Summit 2026 examines MCP, agent-based artificial intelligence, AI architectures, and enterprise governance. Sheng Liang of Obot AI, chief executive officer, joins theCUBE Research hosts to discuss emerging agent-centric architectures and AI infrastructure. Liang draws on experience from Rancher Labs and the cloud-native era to explain how MCP functions as connective tissue for agents, the shift to GPU-driven compute, token-based workflows, and the need for secure production-ready agent systems; they emphasize governance, sandboxing and monitoring at the agent boundary.
The conversation highlights how MCP enables agents to interact with systems and data, unlocking real-world effects. Liang identifies cost and statistical reliability challenges and advocates governance, sandboxing and robust monitoring. Hosts and analysts underscore systems thinking, token consumption as an operational model and practical developer priorities: governance, cost efficiency and validated toolchains.
This discussion addresses enterprise governance for agent deployments, AI infrastructure considerations for GPU-driven workloads, and standards and operational practices that make agent systems production ready.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
MCP Dev Summit - Presented by the Agentic AI Foundation. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for MCP Dev Summit NA 2026
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for MCP Dev Summit NA 2026.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
MCP Dev Summit - Presented by the Agentic AI Foundation. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to MCP Dev Summit - Presented by the Agentic AI Foundation
Please sign in with LinkedIn to continue to MCP Dev Summit - Presented by the Agentic AI Foundation. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Sheng Liang, Obot AI
This session from MCP Dev Summit 2026 examines MCP, agent-based artificial intelligence, AI architectures, and enterprise governance. Sheng Liang of Obot AI, chief executive officer, joins theCUBE Research hosts to discuss emerging agent-centric architectures and AI infrastructure. Liang draws on experience from Rancher Labs and the cloud-native era to explain how MCP functions as connective tissue for agents, the shift to GPU-driven compute, token-based workflows, and the need for secure production-ready agent systems; they emphasize governance, sandboxing and monitoring at the agent boundary.
The conversation highlights how MCP enables agents to interact with systems and data, unlocking real-world effects. Liang identifies cost and statistical reliability challenges and advocates governance, sandboxing and robust monitoring. Hosts and analysts underscore systems thinking, token consumption as an operational model and practical developer priorities: governance, cost efficiency and validated toolchains.
This discussion addresses enterprise governance for agent deployments, AI infrastructure considerations for GPU-driven workloads, and standards and operational practices that make agent systems production ready.
In this interview from MCP Dev Summit 2026, Sheng Liang, chief executive officer of Obot AI, joins theCUBE's John Furrier to discuss how MCP is emerging as the connective tissue that finally gives AI agents the ability to affect real-world change. Liang, a veteran infrastructure entrepreneur and founder of Rancher Labs, frames the current moment as a fundamental shift from deterministic CPU-based computing to statistical, GPU-powered AI systems — a transition orders of magnitude greater in compute than anything that came before. He introduces a compact formul...Read more
exploreKeep Exploring
How has the underlying infrastructure for applications changed with the rise of GPU-powered AI models, and why is MCP becoming a key connective tissue that enables AI to interact with systems and affect the real world?add
How is the shift to GPU- and XPU-dominated AI workloads changing the role of software, databases, APIs, and system orchestration, and what does the new operating-system-like paradigm look like?add
How does MCP differ from a traditional API, and what limitations of AI should be considered when using it?add
How does Obot provide security and governance for enterprises adopting MCP-enabled agents?add
>> Welcome back everyone to theCUBE here at the MCP Dev Summit in New York City. This is a groundbreaking event where the Linux Foundation is announcing their new executive director as well as an intention to have AgentCon. The next big event, we saw KubeCon in the CNCF turned into a massive community. The same thing's happening here in the agent world. Of course, MCPCon is also going to be an event. We're seeing a whole nother level of next gen open source. Sheng Liang is here as the CEO of Obot AI, Obot.ai. It sounds like Robot without the R. Sheng, great to see you.
Sheng Liang
>> Very nice to see you.
John Furrier
>> Been on theCUBE many times. Rancher Labs, sold to SUSE, experienced entrepreneur. At it again.
Sheng Liang
>> Yeah, you can't miss it. It's just been so much fun.
John Furrier
>> You've been in the infrastructure on the CNCF side. This is where most of your CUBE appearances have been on. So you've seen the years of hard work, hardening Kubernetes, all the different projects. That is now the foundation for what we're seeing today, APIs, containers. But now you have a new AI stack emerging with models, MCP, de facto standard, other things going on around it. This foundation represents this next cultural movement.
Sheng Liang
>> That's right.
John Furrier
>> On top of CNCF. Explain what this means from a technology perspective and from a market opportunity perspective.
Sheng Liang
>> Yeah. Yeah. So as you know, I built my career building computing infrastructure, computing platforms for web apps, mobile apps. But now the whole underpinning layer changed. Now, it's all GPU, AI models. So from about a year ago, we saw MCP was becoming a key connective tissue for AI. Without MCP, AI can generate content, it can maybe pull some information and give you some in depth research, but it can't really do anything. With MCP, AI finally gets to interact with my systems, with my data. It can affect change in the real world. So this is just groundbreaking for the first time. I mean, I think for a systems guy, MCP is like a godsend.
John Furrier
>> It's a dream scenario. I agree with you. I've said that many times in theCUBE. I want to unpack that because you look at the fever pitch and the developer community, of course it's obvious. GenAI and with OpenClaw, everyone, you can do so much. There's so much imagination. But software, the software category has been kind of trashed by the SaaSpocalypse narrative, which I'm not for. I don't think that's true. I think bad software dies on its own. But what it speaks to is that it's not the software that you run, it's the system you run. You mentioned it's a systems game.
Sheng Liang
>> Yeah.
John Furrier
>> We are now in an operating system paradigm. Database is a databases. APIs are connectors to agents to flow through. New software like glue needs to abstract and do intelligent things and get work done,, action. This is the new breakthrough.
Sheng Liang
>> Exactly. I think it's not like the software is losing value. It's just the nature of the software change. So we all have to up-level how we function. So what used to be, it used to be that you just manage a bit of CPUs, you do some virtualization and maybe you run a database or two, run some load balancer and you are done. Now, if you compare the amount of compute that happens in the CPU versus the amount of compute that happens in the GPU for AI enabled application, there's no comparison. This is literally orders of magnitude of compute. In terms of power, in terms of operations, happens in the GPU now.
John Furrier
>> Yeah. And then the fact that you have intelligence opportunities to create intelligence, we have for the first time, I would just echo that point and I want to get your reaction to that. And what it means is with GPUs and XPUs and these large scale AI factories, you now have super computing. I mean, forget the Mac Mini, NVIDIA DGX box. Is this big?
Sheng Liang
>> Yeah.
John Furrier
>> Now, it's only $5,000.
Sheng Liang
>> Yeah.
John Furrier
>> You could run at home.
Sheng Liang
>> Yeah.
John Furrier
>> You got the AI factories, which is a massive supercomputer. So now the software has to navigate the complexities of finding the power, knowing where the cycles are for GPU versus compute.
Sheng Liang
>> Yeah.
John Furrier
>> You have tiering of service and Jensen put up the Pareto slide with Vera Rubin that says, "Oh, if you want high performance, use Vera Rubin if you want to use a lower cost."
Sheng Liang
>> Yeah.
John Furrier
>> So you're starting to see that system orchestration, connecting, scheduling.
Sheng Liang
>> Exactly.
John Furrier
>> That's the operating system terms.
Sheng Liang
>> Exactly. So it's not like the ... I mean, people are saying software is losing value. Software companies are losing value. I look at it a little differently. It's not like the software is losing value, but the type of software that does not consume tokens directly are losing value. Because once you start to consume tokens, you function at a different level. You start doing things that traditional software, which you were literally just shifting bits, cannot do anymore, right?
John Furrier
>> Yeah, tokens are the new superpower.
Sheng Liang
>> Exactly.
John Furrier
>> So agents feed on tokens.
Sheng Liang
>> Yeah, exactly.
John Furrier
>> They will eat up tokens for breakfast, lunch, and dinner.
Sheng Liang
>> Exactly.
John Furrier
>> So take me through how you see that playing out and what are you working on now? Talk about your venture, because when you have that token capability, you've got context windows, you've got reasoning. You got all kinds of new things that software can adapt to, make decision making. What are you working on? What's the impact of this?
Sheng Liang
>> So you could almost think about it that way. You could say, "What is an agent?" Agent is an AI app. There's so many ways to define it. But in my mind, agent literally is a mathematically, I would say agent equals to tokens plus MCP. That's how you can understand it. Tokens is what gives agents. It's agency. It's smart. And then MCP is what gives agents ultimately the muscle to do the work. So you really kind of need both.
John Furrier
>> I mean, tokens is like Red Bull. It's like coffee. It's like gas on the tank. It allows the agents to be capable. And so, okay, here's MCP, there's your connective, your doors.
Sheng Liang
>> Exactly. Finally, have an effect in the real world. And this just fundamentally changes how applications are going to be created in the future. You don't really worry about, okay, a system of record followed by some business lodges, wrapped by some user experience. That's how we used to think about these apps. But now, it's literally like the model figures out everything. And then you call the tools, the MCPs to make things happen.
John Furrier
>> Okay. So I have to ask you, because I'm very curious myself, I have my own opinion. I want to get your opinion first. How much does that change you? So take me through your transformation as an entrepreneur, because you've been there, done that. Obviously, we just talked about some of the examples, web apps and cloud. What have you done differently? What's changed in your ideation? How you attack problems? How do you stand up the code, the prototypes? How do you engineer the system? Take me through-
Sheng Liang
>> So the most important thing is as entrepreneurs, we got to solve the problems of today and tomorrow, not solve the problems of yesterday, which are still there, can still be optimized, but are no longer essential. I was just giving an example. No matter how ... We can always optimize the network stack, the scheduling stack, the virtualization overhead. But these things, if you just count how much power, how many cycles is happening, it's just a tiny fraction of the matrix multiplications that happen in the GPU. So you've got to look at how to make what people actually care about today more efficient, more robust, more reliable, more secure. So that's what changed me. So I look at the system today. You got tokens, LLMs figuring out how the system should function. And then you got MCPs that's causing the system to have an effect in the real world. What is the problem? Of course, there are problems about this system is too expensive. So you have NVIDIA, they're pushing the envelope, make it cheaper. But there's also another very big problem, which is, in some sense, this is now statistical computing. Everyone knows that. It's no longer deterministic computing. It's very powerful. But believe it or not, the machines now work more like human. It's no longer 100% reliable. It's not even two nines, not maybe even one nine in some cases. So you got to make sure to have better governance and control. So that's something Obot is focusing on.
John Furrier
>> So pretend that you're the professor of the class and I'm a computer science student. I just graduated four years ago. I know Python, I know C, I know Java and I've been heads down on my team doing work. And all of a sudden I pop up and I'm at a hackathon. I want to build something. I don't really know the tooling. So what would you say to me as a student, to you, the professor, how do I get started? What do I do? I got to change my mindset. You just gave good commentary on that. How do I just get this? What's the tooling? Do I just go to Claude or Cursor? Or how do I know at an MCP? What would I do?
Sheng Liang
>> First of all, what doesn't change is that good computer science, even in a broader sense of liberal arts foundation, right? Math, physics. You need to know how the existing world works. So assume you know that. Then I would say, congratulations. The barrier to get into solving problems is now much lower. It no longer takes three years of apprenticeship plus two years of actual tech leadership to become an experienced engineer who can solve problems. Now, yeah, you plus Cursor or ChromaCode or Obot, whatever, there's so many solutions out there. You can literally be a tech lead on day one. If you can describe the problem, you can solve it. Now, the trick is how do you even know enough to understand what to build, what's worthwhile? That is where your strength as a technologist will shine through.
John Furrier
>> Otherwise, is there like an MCP directory tutorial? How do I know what other servers are out there? Do I create my own? What's the MCP angle on this? Because APIs were easy. REST APIs were very easy to implement. Is MCP as easy? What would be your take on that?
Sheng Liang
>> Yeah. I mean, easy way to understand it is MCP in some sense is actually even easier. So MCP is literally the machine version of API. It's the API optimized design to be consumed by machine as opposed to human, as opposed to developers. That's basically it. So obviously, it's a little different, but it's different in the sense that's actually quite pleasing in the sense that it's actually simpler. And machines are, AI is torn. API is exact. If I just have one misspelling, this is one byte, right? It returns error. MCP will never do that.
John Furrier
>> It's got built in governance.
Sheng Liang
>> Yeah. If a human looks at it, can figure out what's going on, the machine will figure out what's AI will figure out what's going on as well. So this technology is not difficult at all. I think, again, what's more important is to understand what's the actual problem you want to solve, and then understand some of the limitations of the technology. And in AI, the biggest limitation of technology today, number one, is I still believe AI is too expensive just for normal use. Honestly, it's still too expensive compared with traditional computing, which is practically free in comparison. Second is it's just either the ... People used to call it hallucination when they're just generating content, but now when it's affecting the system, it's no longer hallucination. You hallucinate, then you make bad things happen. So now, they call it guardrails or governance or something, security. So you got to make sure the limitations of AI, how much of it you can really trust it. And if you cannot trust it, then what kind of system can you build around it to make it better?
John Furrier
>> Well, great to have you on theCUBE. Talk about what you're working on now, the new venture, Obot.ai, Obot AI as you call it.
Sheng Liang
>> Yeah. So Obot basically provides the security and governance solution for enterprise who want to adopt technologies like MCP. So if you, any organization writing, say writing a new app on AI, what they call an agent, again, the agent will consume tokens. Agent will affect the world through MCP. So what Obot does is it just sits at the boundary of the agent, essentially create an isolated sandbox for the agent. Agent, you can do whatever you want, but if you want to call MCP, go through Obot. If you want to call a MCP, you go through Obot. Then Obot will basically make sure that the agent stays within its lane. It never will do anything that's damaged. If it's something that looks suspicious, it'll alert the user, alert some super admin, alert you to manually approve. So it kind of gives the ... It's basically a governance layer around agents for the enterprise.
John Furrier
>> All right. Well, great to have you on. I love the excitement. I am super excited by agents. I see a generational shift mindset, systems thinking, new ways to be productive, faster time to building and operating.
Sheng Liang
>> Yeah. And again, this is such a great show, because John, I don't know if you know, we created this conference. Obot created this conference where I got-
John Furrier
>> I did not know that.
Sheng Liang
>> Yeah. We created it 10 months ago. The first one, it reminds me of early days of KubeCon. In fact, it was 200 people in-
John Furrier
>> Seattle?
Sheng Liang
>> No, in San Francisco.
John Furrier
>> San Francisco.
Sheng Liang
>> And then last October was 500 people in London. And now, we donated to Linux Foundation in December. And this time you see here, 1,200 people.
John Furrier
>> You're the father of the MCP Dev Summit.
Sheng Liang
>> Exactly. We're the creator of it. Happy to have made a difference.
John Furrier
>> It's a great point. And I like that throwback, because the early days of KubeCon, even before CNCF started, if you remember, Dan Kohn, RIP, great executive director who's no longer with us, an amazing individual, mad respect for what he did. Linux Foundation took KubeCon under its wing because it was just such a small group of people. I was having a beautiful week.
Sheng Liang
>> You remember JJ, Joseph Jackson?
John Furrier
>> Yeah. JJ, he was early player, Google still-
Sheng Liang
>> He created it.
John Furrier
>> All the players, Craig McClure. They were still at Google, by the way.
Sheng Liang
>> Yeah.
John Furrier
>> Burns and everyone else there. But remember, it was tiny, but they were ... Oh geez, they had the vision like you, and so I'm really thankful for what you did.
Sheng Liang
>> Yeah, it's 10 month thing where look at where we are.
John Furrier
>> Yeah, you should have called me up and told me, "Hey, John, come on. I should have known about this. We'll work on that." Great to see you.
Sheng Liang
>> Great to see you. Thank you.
John Furrier
>> Appreciate you
Sheng Liang
>> Thank you.
John Furrier
>> I got an OG here in theCUBE on the cloud native side. The present at creation of this next wave, the Agentic AI Foundation, now a constituted team, a governance board, full community buildup rapidly. So 170 people, MCP, organically built from the community, from entrepreneurs, donating the event. This is what it's all about here with theCUBE here in the open source world, doing our part to be, again, present of creation at another big wave. I'm John Furrier. Gemma Allen here and our entire social media team, getting all the action, sharing as fast as we can to you. Thanks for watching.
John Furrier