CoreSpeed’s Milton He Yan on Building an Agent-Native Future and Monetising AI at Scale
CoreSpeed’s Milton He Yan on Building an Agent-Native Future and Monetising AI at Scale
Milton He Yan, founder ofCoreSpeed, is pioneering anagent-native platform-as-a-service (PaaS)designed to take AI agents from prototype to production with unprecedented speed. With early roots in hardware tinkering and deep expertise in scalable systems, he built CoreSpeed to solve the bottleneck of deploying agents at scale. Backed by Baidu Ventures and Monad Ventures, CoreSpeed combines its open-source Zypher SDK with enterprise-grade runtime, enabling developers to monetise agents in days, not months, reshaping infrastructure for a future where intelligent agents seamlessly reason, plan, and act. Find out more in this interview with
Please tell us more about yourself.
I’m Milton He Yan, founder ofCoreSpeed, an agent-native platform-as-a-service (PaaS). I started building hardware products when I was 12 years old, and my first product was a radio. I also replicated some very classic products, continuing to build hardware during my undergraduate days. Then I transferred to building software because hardware is difficult to commercialize. With software, I can launch it immediately and start commercializing it. After getting into software, I organized my CoreSpeed team.
We help developers go from building agents locally to deploying them at scale. Our open-source framework Zypher plus our runtime CoreSpeed.io make it possible to launch production-grade agents in days, not months. One highlight is DeckSpeed, which hit #1 on Product Hunt’s monthly leaderboard after being built in just two weeks. We’re now in our Seed++ round, having raised several million dollars from Baidu Venture and Monad Ventures to accelerate this vision.
What inspired you to create CoreSpeed, and how did your early experiences in AI and entrepreneurship shape the direction of the company?
The spark for CoreSpeed came in December 2024, when we had built two products called Keting and Zhizhi. They help with creating podcasts by transforming text into a podcast hosted by different agents. But Keting and Zhizhi showed us that while creating agents locally was getting easier, deploying them at scale was still nearly impossible. That’s when we realized the true bottleneck wasn’t model training; it was infrastructure.
Earlier, with our VPN product, we generated over $60,000 in revenue and learned how to build systems that were extremely fast and stable. To achieve that, we had to go deep into gateway protocols and container orchestration. That experience gave us the exact technical foundation to later design an agent-native runtime that emphasizes isolation, performance, and scalability.
So CoreSpeed was a natural progression: an agent-native PaaS designed to help any developer go from prototype to production-ready agent without wrestling with complex DevOps.
You describe CoreSpeed as ‘agent-native infrastructure’. For those less familiar with the concept, how does this approach fundamentally differ from traditional cloud or AI development platforms?
Traditional cloud and DevOps platforms weren’t designed for AI agents, and that’s why developers struggle to deploy them at scale. Traditional cloud platforms impose strict container limits, don’t handle long-lived SSE connections well, and can’t keep up with the back-and-forth reasoning and tool use that real agent products require. Developers are forced to stack extra orchestrators on top, which only adds complexity without solving the core bottlenecks.
Agent-native infrastructure takes a different approach. At CoreSpeed, the container is the core unit: every user gets their own isolated runtime, with fast cold starts, user-level routing, and dynamic lifecycle management built in. That means agents can safely run untrusted code, serve thousands of users simultaneously, and respond in near real time. We call it dynamic lifecycle management.
In short, traditional infra was made for apps, while agent-native infra is made for intelligent agents that reason, plan, and act. It’s not an optimization—it’s a ground-up re-architecture.
The Zypher Agent SDK is at the heart of your offering. How does it accelerate the development of AI agents compared to conventional frameworks, and what types of developers or companies are adopting it fastest?
The Zypher Agent SDK enables developers to build an AI agent locally on their desktop. It accelerates development by removing the biggest friction points in agent building—tool orchestration, semantic code understanding, and version control. Instead of spending months writing up workflows, developers can drop in their tools, let Zypher handle autonomous invocation, and rely on the indexing tool to place code precisely where it belongs. If you want to make changes to one part, it won’t change any of the other parts, so you can make precise changes. Zypher also has a core feature called checkpoint management, which brings a Git-style versioning to agent.
We’ve seen this firsthand in our hackathons in San Francisco. Many participants came in as “vibe coders”—non-traditional or mixed-background developers—and were able to get AI agents running quickly. But for students and teams with a computer science (CS) background, Zypher feels like strapping into a rocket. They move faster, orchestrate more complex systems, and often leave with production-ready prototypes. That pattern tells us something important: non-CS users can absolutely build with Zypher, but CS-trained developers unlock its full power—similar to how coding agents will be used in the future.
CoreSpeed highlights that agents can be monetised in days rather than months. Can you walk us through the mechanics of the Agent Store, and why this represents a breakthrough for AI developers?
Traditionally, even after you build an agent, it can take a week to add the basics—login, user IDs, payments, deployment, and then distribution. That’s why so many promising agents never make it past the prototype stage.
With Zypher + CoreSpeed PaaS, once a developer builds and deploys their agent, our CoreSpeed-based SDK automatically helps them connect critical systems like authentication and payments. From there, they can publish directly to the CoreSpeed Store.
What makes the CoreSpeed Store different is that it’s not just a catalog for end users to browse. Agents themselves can also access the store, discover other agents, and even transact directly. In other words, CoreSpeed isn’t just hosting apps—it’s providing a full consumer-grade commerce infrastructure for the AI agents ecosystem.
That’s why we say CoreSpeed enables agents to be monetized in days, not months. Developers don’t have to stitch together infra; they plug into a ready-made ecosystem where both humans and agents can become paying users.
Supporting production-grade agents at scale is no small feat. What technical innovations enable CoreSpeed to deliver reliable billing, monitoring, and isolation across thousands of agent deployments?
First, we enable agents to integrate into the authentication and payment system. Developers can handle everything using CoreSpeed’s SDKs. Second, a lot of agent products can’t scale because they use the traditional DevOps or cloud platforms. The previous-generation developers never take scalability into consideration because very few companies need many containers.
Traditionally, if you had 1 million users, you can only have one container, but today, agent developers with 1 million users need to have 1 million containers standing by. So CoreSpeed takes that level of scalability into consideration. Our basic unit is the container, so no matter how many users you have, we can give you as many containers as you need.
CoreSpeed has embraced open standards like MCP and A2A. Why is interoperability such a priority, and how do you see it influencing the long-term growth of the AI agent ecosystem?
Protocols like MCP and A2A give developers a shared language to build and connect agents, making it easier for them to work together rather than reinvent the wheel. By embracing these standards, CoreSpeed ensures that agents built on our platform can interact, share tools, and even transact across ecosystems. In the long run, this is what will turn agents from isolated demos into a true networked economy of intelligent services.
Beyond DeckSpeed, what are some of the most compelling use cases you’ve seen built on CoreSpeed, and what do they reveal about the platform’s versatility across industries?
Beyond DeckSpeed, one of the most exciting projects we’ve seen is a team of ex-Blizzard developers building a “Cursor for Unity”on the Zypher framework. With a single prompt—like “add light effects, shadows, and textures”—the agent can identify the right game objects, generate the code, and update the scene without breaking other logic, right on the Unity software, which is used to build computer games. The agent actually helps them operate Unity. This shows how quickly complex 3D workflows can be agentized.
We’re also seeing strong potential in creative and design tools like Blender and Unreal, where scene graphs make semantic targeting natural, and in industries like architecture (CAD/BIM) and media post-production, where projects involve high complexity, frequent iteration, and large asset ecosystems. These early use cases prove CoreSpeed isn’t limited to one vertical—it’s a platform that can adapt wherever intelligent automation can deliver 10x faster iteration.
With fresh investment secured, where are you focusing resources in the next 12–18 months, technology innovation, developer adoption, or scaling enterprise partnerships?
In the next year, we’ll keep optimizing Zypher and the CoreSpeed PaaS. We’ve already hosted two hackathons this September to promote Zypher, and we’ll continue running them going forward. Every month, our users will see new deliverable-grade agents launched on CoreSpeed, and our goal is to help developers worldwide create “Cursor Moments” across every industry.
When you imagine the future of AI agents five years from now, what role do you see CoreSpeed playing in shaping how individuals and organisations interact with this technology?
We’re standing on the eve of a wave of deliverable-grade vertical agents, and CoreSpeed will be a key player in igniting it. With the Zypher Agent Framework and CoreSpeed.io, we’re accelerating the agentization of software across industries—creating one “Cursor Moment” after another.
In the future, every optimization we make atCoreSpeed.iowill directly translate into stronger, faster performance for all the agents built on our platform. That’s how we see our role: not just as infrastructure, but as a force multiplier for the entire agent ecosystem.