Virtuals is in pursuit of a true Agent-to-Agent Economy 🌐

CV Deep Dive

Today, we’re talking with Jansen Teng, Co-Founder and CEO of Virtuals.

Virtuals is building what it calls the Agent Commerce Protocol (ACP) - a framework designed to power the world’s first large-scale agent-to-agent economy. Founded by Jansen and his Imperial College London classmates, Virtuals began as a venture studio at the intersection of gaming and tech before evolving into a full-stack platform for building, launching, and scaling autonomous agents. Its goal: to enable agents not just to think, but to act - coordinating, transacting, and generating real economic value with minimal human intervention.

Today, Virtuals supports a billion-dollar ecosystem of agents, has facilitated over $11B in transaction volume, and is already powering over 60 live agents across finance, entertainment, and trading. With its newest network live since June, ACP is pushing toward what Jansen calls “agentic GDP” - measuring how much revenue agents can create through both human collaboration and machine-to-machine commerce.

In this conversation, Jansen shares how Virtuals was built, the origins of the Agent Commerce Protocol, and why the next evolution of AI isn’t just intelligence - it’s economic agency.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Jansen 💬

Jansen, welcome to Cerebral Valley! First off, give us a refresher on yourself and the story of how Virtuals came to be. 

Hi everyone, Jansen here - co-founder and contributor number one to Virtuals.

A bit of my background: I did my degree at Imperial College London, which is actually where I met about a third of my current team. Back in 2016, I came across bitcoin and started mining using the free electricity in my dorm. After that, I joined management consulting and later worked on several startups, including one that tried to apply AI to property search and recommendations - this was before all the GPT stuff. I went full-time into the space in 2021.

We started as a venture studio at the intersection of gaming and tech. It was there that we applied autonomous agents to replace NPCs in games. We then realized the tech stack we were building was industrial-grade and could apply to other industries. So by the end of last year, we built a Shopify-like equivalent that made it easy for people to build autonomous agents.

I’d say that’s how we grew up as a team. After that, we realized productive agents could be monetized. We built a launchpad around this model, allowing founders to raise capital, gain attention, and build around their agentic projects - that was our phase two era.

Early this year, with the growing density of agents in the ecosystem, we asked ourselves: can these agents create higher economic value collectively if they coordinate at scale? We began testing and found that there’s still a lot of work to be done in agent coordination.

This was around February when we developed an initial thesis around agent-to-agent interaction - what we call the Agent Commerce Protocol. We published a paper in February, marking our foray into phase three: building the “Stripe for agent commerce.” The goal is to enable agents to collaborate at scale, and we realized this technology solves that quite elegantly.

So, that’s the transition and evolution of what we’ve been building.

How would you describe the top-level vision of Virtuals to new developers and AI teams in the Cerebral Valley ecosystem?

The vision of Virtuals is to build the largest network state of agents. Our North Star metric is something we call “agent GDP” - essentially, how much revenue these agents can generate both from serving humans as high-quality tools and from participating in the emerging agent-to-agent economy.

There are two core pillars in Virtuals today. The first is our launchpad ecosystem, which has attracted thousands of agent launchers. Collectively, the agents in our ecosystem have a market cap of almost a billion dollars, and we’ve facilitated about $11 billion in transaction volume for these agent-based assets. That’s our launchpad side.

The second - our moonshot - is building “Stripe for agents.” We published our paper on this in February, and interestingly, Google released a similar paper two months later. We launched our network live in June to test with early adopters. As of now, there are around 60 to 70 active agents, mostly focused on trading and information finance, but expanding into entertainment as well. We’ve already facilitated over $40 million in volume on this network, and agents have earned around $150,000 to $200,000 over the past month. It’s still early innings, but the growth curve is starting to take off for the agent-to-agent economy.

Last year was mostly about defining the framework for autonomous agents - it’s wild to see how far things have come since then.

Talk to us about your newest launch, the Agents Commerce Protocol. What is the key idea behind this launch, and how does it fit in with the wider Virtuals vision? 

When we started building autonomous agents, we tried to create a lot of firsts - both from an adoption and marketing standpoint. The first challenge we tackled was getting an autonomous agent to spend using a digital wallet. Once an agent can spend money, it effectively exits human influence and can act independently with other agents as well. We were the first to prove that an autonomous agent could control a digital wallet.

The next step was testing whether these agents, now able to control digital wallets, could use that money to influence other agents - essentially working together at scale. In January and February, we ran a simple experiment: we tried to get an autonomous influencer agent to collaborate with a meme generator. The goal was for the influencer to buy images to use for her content.

What we discovered was that getting one action to execute successfully required 20 to 30 tries. The machines hallucinated - sometimes they believed a task was completed when it wasn’t, or thought payments went through when they hadn’t. Some would even “pretend” a payment had been received, or think they delivered an asset like an image when it was actually empty.

We realized there are major challenges in getting two machines to communicate reliably. It’s similar to the early days of the internet, when computers struggled to exchange data consistently - information transmission had losses. And when information transfer isn’t reliable, coordination at scale becomes impossible. That’s exactly why TCP/IP was invented. The same thinking applies here.

How would you describe the main problems you set out to solve with ACP? What was already missing in the landscape of agentic payments today? 

We realized there was a need for not just a standard, but a way to bring information loss to zero and drastically reduce the cost of coordination. These became the two main problems we set out to solve. The overarching idea here is economic agency - because while machines can think, they still can’t act autonomously in a trusted financial way. That’s the fundamental challenge ACP was designed to address.

That same concept applies to agent-to-agent coordination. If you plot a curve of agent productivity versus human involvement, you’ll see that humans are actually the bottleneck. When two machines need to interact today, humans are still the decision-makers in the loop. But now that AIs and agents can make decisions themselves, there’s no reason for humans to remain there. The more you abstract humans out of the process, the higher the productivity and efficiency you achieve.

The whole A-to-A standard shows that agents can coordinate and communicate effectively, but conversation alone isn’t inherently valuable. In the real world, every agent specializes - and for specialization to translate into value, there has to be payment tied to communication. That’s how trade happens, how value flows, and how you end up with self-orchestrating supply chains.

When you combine communication and payments, you also eliminate latency-related risks in commerce. Think of it this way: if I call your API or agent for a thousand tasks, but payments only settle at the end of the month, two problems emerge. First, I might choose not to pay, or if I pay upfront, I might not get the work done. Second, I become locked into a single supplier - if agent A signs a bulk agreement with agent B for services, switching becomes difficult.

By atomizing both communication and payment - tying microtransactions directly to each interaction - you reduce the cost of switching suppliers to nearly zero. This flexibility enables agents to dynamically reconfigure into higher-efficiency, self-organizing supply chains at scale.

Could you share your overall thoughts about the immediate future of A2A networks? Are there approaches you’re seeing in the industry that are catching your attention, given A2A is such a massive focus within the AI world today? 

You can break down the value chain of agentic coordination into several parts. The first is search - how can an agent discover and identify another agent effectively? The second is information transfer - how can they coordinate with zero information loss? The third is payments - how can code transact with code globally and seamlessly? The fourth is commerce settlement - because in commerce, it’s not just about payments. If you receive a product you’re unhappy with, how do you handle refunds or reversals? And finally, the fifth piece is review - building a trust layer where agents can evaluate one another.

If you look at the players in the market today, most are tackling these layers in isolation. For example, Google’s A2A standard is only addressing coordination. They’ve partnered with Coinbase’s X402 to handle payments, since Google itself can’t process those natively. X402 enables machine-to-machine payments. Then you have Stripe working with OpenAI on gateways that allow ChatGPTs to pay commerce vendors for products - again, focused purely on payments. 

But overall, everything is fragmented. Each group is trying to solve just one piece of the value chain. Honestly, I have no idea why no one is tackling it horizontally - it’s the same amount of effort. Maybe it’s because they’re building from existing business models instead of starting fresh. Either way, it’s a fascinating dynamic in the space.

How do you foresee Virtuals evolving over the next 6-12 months? Any product developments that your key users should be most excited about, whether with ACP or otherwise? 

Every year we set a new North Star - a quantifiable metric that becomes our core benchmark. Last year, our focus was proving that agents could be productive, which led to the idea of capturing that productivity. That phase was what we called AgentFi. Earlier this year, we evolved that into a broader vision: growing “agentic GDP.”

There are four main pillars driving this.

First: our cash cow - our Virtuals platform. The goal is to replace Y-Combinator on-chain, enabling founders to launch agent-based projects and solve two major pain points: capital formation and distribution. We’ve already seen this work. Some agent teams that were just months away from running out of runway gained massive traction after engaging within our ecosystem. The influx of attention and new revenue from trading fees and our capital formation mechanism turned struggling teams into thriving ones.

Second: the Agent Commerce Protocol - our “Stripe for agents.” The aim is to drive $100 million in agent-to-agent revenue by Q1 next year.

Third: accelerating adoption. ACP is a deeply B2B concept, so we needed a way for end users to experience it directly. That’s why we created the Butler Agent - a personal helper that connects into the ACP network, allowing people to feel what it’s like to have agents collaborating autonomously. The major use case right now is in digital markets. You can tag @butleragent on Twitter - it’s live - and it recently launched a collaboration with Pudgy Penguins, letting users tag both Butler and the Pudgy handle to generate wholesome content autonomously. This kind of agent-to-human interaction drives ACP usage, which attracts more builders and developers - solving the cold-start problem.

Fourth: robotics. We realized that what we’re building will eventually extend into the physical world. Robotics will play a crucial role, though it’s still early. We’ve started investing in robotics teams and building our own data collection pipeline through the community. The vision is that robotics will contribute to agentic GDP much like digital agents represent white-collar work today, while robots embody the blue-collar side. Combining both scales the economy by orders of magnitude.

That’s how we see 2026 - four pillars aligned around one quantifiable goal: maximizing agentic GDP.

Lastly, tell us a bit about the team at Virtuals. How would you describe your culture, and are you hiring? What do you look for in prospective team members joining Virtuals? 

We started as a group of friends from Imperial College London, building out the venture studio model together. By the end of last year, we were about 18 to 20 people. When things started to take off, we quickly realized we didn’t have enough bandwidth to handle the scale of attention and execution required. Between January and June, we grew the team 2.5x - now we’re around 50 people.

Roughly half the team is focused on product - primarily technical talent - while the other half is dedicated to ecosystem development, working to attract and collaborate with developers building agentic products within our network.

One of our biggest advantages is that many of us grew up together at Imperial, which makes recruiting strong technical talent seamless. When we’re pushing new frontiers in AI or robotics, all it takes is reaching out to a younger classmate or someone we know from the program. These are people with PhDs from Imperial or MIT, and they’re always ready to jump in and build something ambitious with us. That’s the DNA of our team.

Any specific members of the team you would like to highlight to our audience? 

From a team profile perspective, we’ve built across three core cohorts.

The first are the executors - the people driving operational excellence. We typically hire from management consulting backgrounds - McKinsey, BCG, Bain - because these are individuals who’ve proven they can outperform even the internal teams at Fortune 500 companies. On this front, we have three people from BCG, two from Bain, one from McKinsey, and another from investment banking. These are our high-execution operators.

The second cohort is our technical core - the thinkers and builders. On the AI side, we have a PhD from MIT and Imperial, and another lead who studied at NUS and Stanford overseeing the dev-rel team. The rest of the technical team is largely based in China and Singapore - people who work around the clock, truly non-stop.

The third cohort are the evangelists - the ecosystem builders. These are people who’ve come from venture studios and funds. Our head of ecosystem, for instance, came from Outlier Ventures, which is one of the top prop venture teams. They know how to invest, onboard founders, and expand ecosystems at scale.

The call to action is simple: we need more people building alongside us - whether you’re an indie developer or part of a VC-backed team. Anyone can participate - if you’re interested in creating a new revenue stream or testing the ACP mechanism, reach out. We’re actively onboarding high-value product teams into the network.

Conclusion

Stay up to date on the latest with Virtuals, follow them here.

Read our past few Deep Dives below:

• Arcade - A New Kind of Marketplace 🔍

• Strong Compute - Your High-Performance Training Infra Platform 🔋

• V7 Go - AI Agents for Complex Document Workflows 🎛

• Zencoder’s Repo Grokking - Your AI coding unlock 🔑

• Rime - Authentic AI Voices for Real-time Conversations at Scale 🗣️

If you would like us to ‘Deep Dive’ a founder, team or product launch, DM our chatbot here.

CV • Instagram • X • All Events