Original · Unique Research · 2026-03-23 · Shanghai
Editor's note: This complete English edition retains the original author's first-person commentary and the interview's March 2026, pre-summit context. Descriptions of Buda's future pricing and AI organization, and the forecast of 5 humans working with 500 AI agents, are plans and opinions rather than verified deployments or guaranteed outcomes; the original caveat about timing is retained. Founder credentials, user scale, funding and the customer-service anecdote have not been independently verified here. Regional comparisons describe the source's perspective, not every person or business in those markets. “Tutu Yangxia” romanizes 兔兔养虾; an official English name has not been confirmed. This article should not be read as evidence that the 50-agent setup reported in an earlier interview expanded to 500 agents.
Unique Awards · Guest Interview
He No Longer Wants to Sell Software:
What He Wants to Build Is a Company's 500 AI Employees
Buda.im Founder Kelly: Could AI Rewrite the Very Idea of a Company?
Lately, one feeling has been growing stronger for me:
Many people talk about AI, but what they are actually thinking about is still software.
Many people talk about agents, but still use them as tools.
In other words, on the surface the whole industry has entered a new stage, yet most people's way of understanding it remains in the previous generation.
That is why I have recently been particularly keen to talk more with founders who have already begun thinking about companies in a new way. Some, you discover, are no longer satisfied with having AI write something, look up information or run a workflow for them.
"
What they really want to do is something more radical: What happens if most of a company's employees are no longer human?
Before this Hangzhou event, I spoke with Kelly. He is the founder of Buda.im and Tutu Yangxia, as well as Vika / APITable, and previously served as HEYTEA's founding CTO.
Honestly, impressive though those titles are, they are not what stayed with me. What I remember is his entrepreneurial instinct: not "I want to make a feature more powerful," nor "I want to enter a hot market," but a genuine attempt to answer a more fundamental question: How should companies be organized in the future?
That is also why I think he is worth listening to at the summit. He is no longer talking only about international expansion, agents or AI tools. He is touching a bigger question: Could AI rewrite the very idea of a company?
From Vika to Buda: Not a Pivot, but a Questioning of the Software Era Itself
As Kelly described his move from Vika to Buda, what struck me most was that this was not a story about finding a new direction. It was about a change in his underlying understanding.
Vika addressed a classic software problem: how to help human teams collaborate more efficiently. There was nothing wrong with that, and it had already done it well. It had served over a million users, its open-source project attracted considerable attention, and it had raised a respectable amount of funding. By startup standards, that is a very creditable path.
But after AI emerged, he began thinking about something else: What if AI, rather than humans, did most of the actual work in a team?
That can sound like an abstract concept. But he did not leave it there. Around the Lunar New Year, he began using OpenClaw intensively. Through that experience, he felt something very concretely for the first time: One AI agent and a group of AI agents are entirely different things.
You can think of a single AI as a tool. Asking it to write, research or complete a task is still tool-oriented thinking. But things change when you deploy a group of AI agents according to a company's organizational structure, give them distinct roles, have them collaborate and connect their work, and even develop a kind of team behavior. It is no longer just a helper. It looks more like the beginnings of an organization.
I think many people have not yet truly grasped this. Plenty say they are using AI, when in fact they have simply acquired a few smart tools. But perhaps that is not where the real change lies. The real change is this: Once multiple agents are organized, AI begins to look less like a tool and more like an organization.
That was the starting point for Buda.im. It is not a simple wrapper product or a middleware layer that merely connects several models. Kelly wants to build a system in which AI employees can persist over time, continually spawn new ones, collaborate and evolve.
He says Buda is more like an office building. I like that analogy. He is not selling an isolated capability. He is trying to build a place inhabited not by people, but by a group of AI employees.
He Says, "Software Is Dying"
It sounds exaggerated, but I think he is more than half right. Over the past few years, we have become far too accustomed to understanding AI through efficiency gains. Efficiency matters, of course, but it is also a term that can easily make the change sound smaller than it is.
If AI only improves efficiency, it is essentially patching the existing software system. The workflows, forms, menus, back ends and switching between systems all remain; things are just a little faster. Kelly's judgment is more radical. He says software is dying.
On first hearing that, many people might instinctively find it too absolute. But think it through and you realize that the idea is both unsettling and quite reasonable. Most of us do not actually like software. No one genuinely loves expense systems, CRM, ticket workflows, back-end configuration or permissions menus. People use software because they need to get work done. Put plainly, we do not enjoy operating software; we have been forced to learn how.
So if AI is beginning to understand language, the more natural way to work in the future should not be to study where to click in one application, what to enter in another system, or which menu level contains a feature. It should be this: I state what I need, and AI gets it done.
That is a very large shift. At the heart of the software era was the idea that people adapt to systems. In the AI era, what is more likely to happen is that systems begin adapting to people.
What changes underneath is not just the interaction model, but the assumptions on which the entire software industry rests. If language itself can become the entry point, tasks can be understood directly, and workflows can be orchestrated dynamically, then a large part of the moats built from interfaces, modules, processes and training really could be eroded. So I understand why Kelly would move from building collaboration software to building an AI organizational system. Once you truly believe language will become a new operating system, it is hard to remain satisfied with making software easier to use. You begin wanting to build new infrastructure.
The Biggest Divide Today May Be Between Using AI and Managing It
In conversations with founders, investors and product people, I notice a clear difference: Some are using AI. Others are already managing AI. That sounds like a change of just one word, but behind it lie almost two different eras.
The former still follows a user's logic: I ask a question. I generate some text. I ask it to make a spreadsheet. I ask it to write an email. The latter is already moving into organizational logic: I set goals. I break down tasks. I assign different agents different roles. I check results. I exercise judgment. I adjust the mechanisms.
Kelly is clearly in the second category now. That is why Buda's target users are well defined: not everyone, but people who particularly want to amplify themselves into a team. Whether highly leveraged solo operators or founders of small and medium-sized businesses, they are pursuing essentially the same thing: organizing greater productive capacity with fewer humans.
There is an interesting business logic here. Kelly says their future pricing will be per human. I find that an especially clever judgment. In the AI era, AI employees and computing capacity can, in theory, be replicated continually. What may actually be scarce is the person who gives instructions, makes the final judgment and bears responsibility.
Previously, a major company cost was hiring more people. In the future, the company's core resource may become fewer but stronger human managers. This is not simply cutting costs and improving efficiency. It looks more like a reconstruction of the company's resource structure.
Why Many Companies Lose Overseas Not on Product, but Because They Cannot Find the Right People
The roundtable Kelly will attend concerns regional market entry: differentiated strategies for Japan, Europe and the US, and the Middle East. But what makes him worth listening to, I think, is not necessarily the tactical detail of any one market. It is his understanding of what local execution really means. He put it very plainly: Many business problems ultimately come down to whether you can find the right people.
That may sound unsophisticated rather than like an elegant methodology. But precisely because it is not elegant, it rings true. People discussing international expansion often prefer a high-level view: strategy, regions, timing, brand, localization, channels and compliance. None of that is wrong. But many businesses do not fail on a PPT slide. They fail in very concrete circumstances: They cannot find anyone. They cannot establish connections. Nobody handles problems when they arise. Customer service is slow. Partners are unreliable. There is no practical foothold for getting things moving locally.
Kelly gave an example: An investor had a problem with China Mobile Hong Kong mobile data and wanted to change plans. They could not reach anyone by phone. WhatsApp customer service required waiting in a queue, and when their turn came, they no longer had time. In the end, they had to travel to a store themselves. This is hardly some grand business question. Yet this is often exactly how the business world gets stuck. If it happens in Hong Kong, Japan, the Middle East and Europe will only be more complicated.
Why, then, do so many supposedly international-expansion services leave customers feeling little benefit? Because they offer advice, not a practical foothold; plans, not progress; frameworks for understanding, not the people and paths that solve problems. What Kelly now wants to do is gradually entrust finding people, collaboration, execution and localized progress to a team of digital employees. Others sell tools. They want to deliver the beginnings of a team that can keep working within the local language and cultural context.
AI Can Replace Many Parts of the Work, but It Will Also Create a New Role: AI Administrator
When asked what AI has actually replaced, Kelly answers directly. He believes three types of work are the first to be visibly replaced: content, research and execution. Market research, localized content generation, customer service and operational execution can already be accelerated substantially. Many tasks that once required a small team taking turns can now get underway with a few agents and one person who knows how to manage them.
But an easily overlooked point matters here: AI replacing work does not mean the work disappears. Often, it is simply redistributed. Previously, you did it yourself. Now you must define goals, break down tasks, check quality, correct direction and manage a group of AI systems. A new role emerges: the AI administrator.
I think roles like this will become increasingly common. Many people who believe they are learning tools today may eventually discover that they are learning how to supervise a digital team. This is another feeling that has been growing stronger for me: What AI ultimately changes may not be one job, but the division of labor throughout an organization. You are no longer the executor yourself. You increasingly become a coordinator, a verifier and a decision-maker. That change will be enormous for individuals and companies alike.
The Hard Part Is Not Translating Language, but Understanding Culture
Kelly also made a point I strongly agree with. When expanding abroad, many companies instinctively assume localization is mainly a language problem. Translate the website, revise the copy, create another version of the advertising assets, and it seems nearly done. But anyone who has genuinely worked in overseas markets knows language is only the most superficial layer. The real difficulty is culture: how trust is built, how relationship networks operate, what users truly care about, and how people in a particular country or region actually make decisions.
Japan places great weight on long-term trust. Many opportunities in the Middle East arise through personal networks. Europe and the US have very different privacy and compliance requirements again. None of this can be solved by AI automatically generating a few passages of localized content. That is why I think one of Kelly's judgments is worth remembering: AI handles efficiency; humans handle judgment.
The value of that statement is that it neither glorifies nor underestimates AI. AI can first give you an 80-point version. That is already impressive. But the final 20 points often determine success or failure. And those 20 points still depend heavily on humans' long-term understanding of the market.
When Resources Are Limited, Wanting Everything Is More Dangerous Than Moving Slowly
When companies first expand abroad, many instinctively want to tackle Japan, Europe and the US, and the Middle East all at once. If they are going global anyway, why not do it together? It sounds bold. In practice, it often ends badly. What look like three markets on a map are three entirely different sets of operating logic in reality. With limited resources, the most rational approach is usually not to advance on three fronts, but to win one first.
Kelly proposed a simple prioritization framework that I find useful: first, the strength of market demand; second, the difficulty of entry; third, the fit with available resources. Is there genuine demand for your product in that market? Will cultural, policy or competitive barriers be too high? Does your current team have the necessary language capabilities, resources and experience? Looking at these three dimensions together usually makes priorities much clearer. Many companies do not lose because they cannot see opportunities. They lose because they want too much, try everything, and never establish a foothold anywhere.
Will General-Purpose Large Models Eliminate Products Like This Too?
This is one of the questions I most wanted to put to him, because it is so practical. If models such as ChatGPT and Claude are already so powerful, why would customers not just build something themselves? Why do they still need a system like Buda?
Kelly's answer is candid. He says that if he were still on the customer side today, he would not buy many traditional software products either. AI has lowered the barrier to doing it yourself dramatically. It costs less, adapts quickly and lets the results stay in-house. For many traditional software companies, the impact really is devastating.
But he also stresses a point: A smart individual model does not mean multiple agents can form an organization. One person with a general-purpose large model can do many things. But getting multiple agents to divide work, collaborate, cross-check one another, share long-term memory and complete an end-to-end process is not a capability that simply appears on its own. It is more like a problem of enterprise digital infrastructure. The issue is not whether you have one intelligent mind, but how to make a group of intelligent minds collaborate. I think this is precisely a divide worth watching in the period ahead. Individual intelligence will continue to become more competitive. But what can genuinely sustain commercial value may increasingly shift toward organizational intelligence.
Why I Think This Conversation Is Worth Hearing
Kelly is betting not just on a product, but on a new form of company. He offered a final prediction: By the end of 2026, the era of measuring company size by employee headcount will come under rapid pressure. The most efficient international-expansion strike teams of the future might consist of just 5 humans and 500 AI special-operations soldiers. That is certainly a bold claim. Things may not unfold at exactly that pace.
But to me, its real value lies not in the numbers. It lies in forcing us to rethink assumptions we once took for granted: What is a team? What is scale? What is organizational capability? What counts as infrastructure for international expansion? We used to believe that growing a company required continually hiring people, building departments, adding systems and establishing processes. Now another possibility is emerging: a small core of humans, many AI employees that can be replicated, coordinated and made to collaborate, and a system that can truly organize those agents.
If that path proves viable, the deepest impact of AI on business may be impossible to capture with the phrase efficiency gains. It is rewriting companies—and globalization. That is why I think Kelly's panel at this summit is not only for people interested in differentiated strategies for Japan, Europe and the US, and the Middle East. It is even more relevant to those asking how they should build teams and organizations in the future, and use AI to amplify themselves. Sometimes a person's value is not in how many answers they give you, but in helping you ask a more ambitious question first. That is what Kelly is doing now.
Selected Interview Q&A
Q1: How do you define Buda.im today?
A: Kelly puts it directly: It is not a single-purpose AI tool, but a collaboration platform for multiple AI agents. It gives you not a tool, but a team of digital employees that can work together.
Q2: What fundamentally changed from Vika to Buda?
A: Previously, the problem was how human teams collaborate. Now, the question is how an organization operates if its employees become AI. This is therefore not a pivot in the ordinary sense; the underlying problem has changed.
Q3: Why does he say software is dying?
A: Most people do not actually like software; they are compelled to use it to get work done. As AI begins to understand language, the more natural way to work in the future may be to state what you need and let AI complete the task, rather than learn software. Language may become the new operating system.
Q4: Which parts of international expansion has AI already replaced?
A: The most visible areas today are content, research and execution. Market research, localized content generation, customer service and operational execution, for example, can already be accelerated substantially.
Q5: What is hardest for AI to replace today?
A: Cultural understanding and business judgment. Language can be translated and content generated, but people still need to judge the trust mechanisms, relationship structures and compliance logic underlying each market.
Q6: Why do many companies fail overseas for reasons other than poor products?
A: Many have entered a market without becoming part of it. Signing agents, holding launch events and translating a website are not difficult. The hard parts are building trust over time, obtaining continuous feedback and genuinely understanding local users.
Q7: Should companies pursue several regions at once if resources are limited?
A: Kelly's advice is to win one market first. Prioritize using three factors: the strength of market demand, the difficulty of entry and the fit with available resources. Do not start by wanting everything.
Q8: Will general-purpose large models directly replace products like this?
A: They will severely disrupt much traditional software, but may not naturally solve organizational collaboration. A smart individual model does not mean multiple agents will naturally form a team. Buda aims to solve more infrastructure-like problems: collaboration, memory, division of labor, verification and end-to-end execution.
Q9: Who is Buda best suited to?
A: Two groups: highly leveraged solo operators who want to amplify themselves into a team, and founders of small and medium-sized businesses who want to organize greater productive capacity with fewer humans.
Q10: What breakthrough matters most to Kelly in 2026?
A: Not a particular feature or region, but whether more people can truly accept that AI is not just a tool—it can become part of a company.
Q11: Where does he see the greatest new opportunity?
A: Digitally native micro-multinationals: small core human teams using AI leverage to do business worldwide.
Q12: If someone at the event wants to collaborate, what is the best starting point?
A: Do not start by discussing a huge project. Take a specific pain point in a real target market and run the smallest pilot that can be validated within a week, to see whether AI employees can first replace part of the workflow.
This document is original content from Unique Research.