NUPIAO
NUPIAONUPIAO
As foundation models get smarter, reasoning costs keep dropping, and agent technology matures, enterprise AI vendors are racing to push their products beyond generic assistants into the core of business operations.
Kai-Fu Lee, founder and CEO of Zero One, sees this as the dawn of a new phase for enterprise AI. He sums it up as moving from “boosting execution efficiency” to “improving decision quality.”
On July 7, Zero One rolled out a suite of AI products aimed squarely at the C-suite — “Boss AI,” “Sales Champion AI,” and “Investment Officer AI” — each tailored to critical decision scenarios in business management, sales, and investment. The company also launched “Wance,” an enterprise decision platform that acts as the underlying engine, stitching together corporate data, ontology modeling, multi-agent collaboration, and execution loops.
Think of them as Windows and Office — one is the system, the other the application. They’re priced separately. Zero One claims that after deploying these tools internally, its own order volume grew fivefold and sales conversion rates doubled.
In Zero One’s view, most enterprise AI applications over the past two years have been little more than efficiency boosters — helping staff write documents, generate code, summarize meetings, or build knowledge bases. Sure, they save time, but they rarely move the needle on revenue, profit, or cash flow.
But the next battleground for enterprise AI isn’t just about streamlining tasks; it’s about diving into business scenarios that directly shape financial performance — things like operations management, revenue growth, risk control, and investment decisions.
That’s why Zero One is positioning its products for the “No.1 position” — the CEO and top business leaders — trying to push AI from a supporting role right into the heart of business performance. The company’s competitive set is shifting from office software and generic agents toward the enterprise decision software market. It now sees U.S. data analytics firm Palantir as a much more direct peer.
Over the past year, Zero One has been redefining its own ecosystem. Instead of talking up model capabilities, Kai-Fu Lee now talks more about Ontology (making AI systematically and structurally understand a company), dynamic context, multi-agent collaboration, FDE (Frontline Delivery Engineers), and enterprise AI operating systems.
All these concepts point to one idea: models handle reasoning, but the real competitive moat lies in a company’s business knowledge, organizational structure, data governance, and the closed-loop decision-making that runs continuously. He stresses that in the future, enterprise clients won’t bother asking which company built the underlying model — large models will be as invisible and ubiquitous as electricity or tap water.
This conviction is also reshaping Zero One’s own trajectory. In a post-launch media Q&A, Kai-Fu Lee mentioned that the company is pushing forward a partnership system and a DRI (Directly Responsible Individual) mechanism, aiming to reduce the organization’s reliance on any single person.
He admitted that while he has had an advantage in reaching out to CEOs, over the long haul the company must turn that resource into scalable products, platforms, and delivery systems — not just founder-driven influence.
On the global stage, Zero One sees plenty of opportunity beyond North America. Many countries still lack mature enterprise AI services. Top U.S. model companies focus mainly on the U.S., Europe, and Japan, but Chinese firms — with stronger engineering delivery capabilities, multilingual service chops, and more flexible business models — can carve out an edge in markets like the Belt and Road region.
When it comes to commercialization, Kai-Fu Lee believes there won’t be just one model. Token-based APIs will remain a key infrastructure play in the AI era. But for complex, high-stakes scenarios that directly impact business results, simply calling a model won’t cut it — you still need long-term delivery, deep customization, and ongoing operations.
“For those core enterprise scenarios where you can’t tolerate an 85% score and must hit 99% — the high-value, high-difficulty, high-focus ones — you have to adopt something like Palantir or our own product system,” Kai-Fu Lee said. “These two business models are not in conflict.”
Below is the edited transcript of our interview with Zero One:
Media: As Zero One pushes “Boss AI” and other No.1-position initiatives, how do you view the opportunity for AI rollouts in non-U.S. markets? What unique competitive advantages do Chinese AI companies have there?
Kai-Fu Lee: When we promote products like Boss AI overseas, we’ve found that many countries haven’t even figured out how to use agents properly. They need a lot of help with agent building, governance, and evaluation.
Top U.S. model companies are currently focused on markets that pay quickly and are culturally and linguistically familiar — the U.S., Europe, Japan. A huge chunk of the world isn’t being well served by them. And those markets also need deep FDE (Frontline Delivery Engineer) engagement, which U.S. giants are usually reluctant to send lots of engineers for.
This is a massive opportunity for Chinese companies, especially along the Belt and Road. Of course, to capture this market, there are two main hurdles: First, whether you can reach the top decision-makers in these countries. That’s exactly Zero One’s strength. Second, multilingual fluency.We’re building capabilities in English, Kazakh, Serbian, and other languages.
On the data side, we will neither take Chinese data out nor bring foreign data in. That’s a non-negotiable bottom line; otherwise, we wouldn’t win any deals.
Media: A while back, Zero One’s internal letter mentioned several heavy-duty incentive measures — a partnership plan, stock options, and CEO special incentives. How are those progressing? Do you feel they’ve boosted morale?
Kai-Fu Lee: Frankly, morale has been pretty high for the past six to nine months. These measures might add a little extra, but I’m more focused on making sure truly outstanding people get the rewards they deserve, rather than just lifting spirits.
Right now, they’re all on track. I just approved the first batch of the new option plan. As for partners, because we’re still a relatively young and lean company, we need more time to observe deeply. The first batch of partners will definitely be announced within the next six to nine months.
Next, we’re looking for DRIs (Directly Responsible Individuals). We’ve already assigned DRIs on some projects internally (just named one yesterday). We’ll keep flattening the organization, giving DRIs and partners more authority to run their businesses independently.
This is what every modern company in the AI era has to do. Since Zero One advocates this methodology, we have to practice what we preach; otherwise, clients won’t buy it.
Media: In practice, how do you make AI products “one thousand enterprises, one thousand faces” — offering differentiated decision support for different industries and growth stages?
Kai-Fu Lee: Zero One’s product suite is already highly complete in terms of productization. But even so, it’s still not “plug-and-play.” We have to send FDE teams to help clients build their unique Ontology and do technical adaptation.
That also means our product isn’t suitable for SMEs — it needs a fairly large enterprise to bear the cost. Once we deploy FDEs, the technology and platform can generate real value.
Of course, different industries have different rollout difficulty. Purely digital sectors (like investment) are faster; industries with heavy traditional manufacturing take longer, but the value created is universal.
Media: While pushing “No.1 Position” projects in many large and mid-sized enterprises, what core business methodology has Zero One distilled?
Kai-Fu Lee: The first lesson is that AI transformation must be led by the No.1 position — there’s no second choice. Second, the No.1 position focuses on core scenarios that are cross-functional and cross-departmental, which is why only the No.1 can pull it off.
Third, many middle managers and frontline employees actually have some resistance, fear, or anxiety about AI. If you hand AI transformation to lower-level teams, some people will worry about their jobs or status because they don’t know how to manage AI. That’s when the No.1’s rallying power becomes huge — if the No.1 uses AI openly and says “I use it every day, you have to too,” it makes a big difference.
On the technical path, we need to build a decision hub that includes both a model layer and various functional modules. Through that hub, we can develop a relatively reusable product.
Media: If CEOs use AI to make decisions, will that put new pressure on employees? Before, they only had to communicate with the CEO; now they might have to explain themselves to AI’s judgment. Do you think the relationship will get more complicated, and communication costs higher?
Kai-Fu Lee: No, because our AI doesn’t directly interrogate employees. It quietly collects data from multiple sources, objectively raises judgments and suggestions, and the final decision power stays firmly in the CEO’s hands.
Since I started using Boss AI myself, I still meet employees just as often. But the questions I ask have become more relevant, more precise, and sometimes sharper — which actually helps them do their jobs better.
Media: In your earlier internal letter, you said Zero One aims to become “China’s first profitable AI 2.0 company” and plans an IPO by 2027. But deep co-creation with large enterprises and government clients usually means high contract values but long deployment cycles and slow cash collection. With the launch of Wance AI, how will Zero One scale up and reach breakeven this year?
Kai-Fu Lee: I need to clarify — what I mentioned was hoping to achieve profitability “in a certain quarter next year,” not this year.
You’ve noticed the gap between big contract amounts (order value) and actual recognized revenue. Many of our large deals, both domestic and international, use milestone-based payments. So the revenue from big orders gets recognized gradually over one to three years.
We also know some industries see contract cancellations, but each of our deals delivers critical technology, and every year we iterate based on the previous phase’s delivery. So far, client satisfaction has been very high.
Media: Over the past year, you’ve visited many clients to push AI adoption. Will you keep doing that? How do you choose who to visit? And what common pitfalls have you seen in enterprise AI transformation? Any differences between domestic and overseas clients?
Kai-Fu Lee:Over the years, I’ve always been willing to be on the front line, and that won’t change. But our team is excellent too. We have multiple approaches to opening doors: sometimes we start with the CEO or CIO, and the team makes initial contact; I only step in when necessary. For certain CEO-focused talks, I’ll participate more actively.
Plus, many CEOs reach out to us proactively after seeing our announcements — like today’s product launch. In those cases, if they need me, I’ll definitely go, but most of the time the team handles client relationships just fine.
Media: Compared to competitors, what’s Zero One’s hardest-to-replicate moat? Is it your personal reputation and network, the accumulated insights from talking to CEOs, or the model and technology itself?
Kai-Fu Lee: In the early days of doing “No.1 Position” projects, I did have some unique advantages in reaching top decision-makers. But as we dug deeper with these users and completed a few project-based engagements, we started productizing the services. During that process, we gained a deep understanding of common client needs and built solid technical expertise in areas like Ontology.
I believe our moat is a “systems engineering” advantage.It’s like Microsoft’s Windows system. If you ask what its advantage is, it’s not necessarily one single technology or Bill Gates’ personal edge. Zero One wants to achieve the same: by broadly engaging No.1 positions, understanding common needs deeply, and combining that with cutting-edge tech, we form a systemic advantage.
Media: Does Zero One’s “No.1 Position” strategy rely heavily on your personal industry prestige and network to open CEO doors? If one day you stop personally meeting those hundred-plus CEOs, can this model still work?
Kai-Fu Lee: As CEO, one of my core duties is to ensure the company’s long-term sustainability. So making the company highly dependent on me is by no means something I’m proud of — on the contrary, I’ve been pushing hard to build the company’s ability to run continuously.
First, I’m very healthy and in my prime, so you can expect my long-term service.From a business standpoint, more and more of our big No.1-position deals are being won by the team itself, not by me, and that proportion is growing fast.
Second, my role in the “No.1 Position” initiative is mainly about opening doors early and maintaining top-tier relationships; it doesn’t take up much of my daily time.
Third, since I started using Boss AI, it has freed up a ton of internal management time, giving me plenty of bandwidth to handle high-end client engagement.
Finally, our target client base is pretty clear: globally, there are maybe two to three thousand mid-to-large enterprises fit for our high-barrier product. In the past two years, we’ve already touched five to six hundred of them. Over the next two years, having the team cover the rest is absolutely no problem.
Media: Palantir’s CEO recently criticized OpenAI and Anthropic, claiming that a business model based purely on selling tokens has a fundamental value flaw. What do you think about the commercial value of tokens? For enterprise deployment, is it better to sell tokens or deep, packaged enterprise solutions?
Kai-Fu Lee: Tokens definitely have huge value — otherwise so many top companies wouldn’t bet on them as their core business model, and there are indeed massive numbers of developers and customers buying them. They offer a low-barrier, pay-as-you-go approach.
But the limitation is this: when a company faces problems that are more unique, carry higher core business stakes, and involve more complex decision-making processes, a single generic token interface has a harder time solving them. For some enterprises, a generic token API with a lightweight application is enough — that’s a perfectly logical, large-scale business model.But what Palantir pointed out is that for those core enterprise scenarios where you can’t tolerate an 85% score and must hit 99% — the high-value, high-difficulty, high-focus ones — you have to adopt something like Palantir or our own product system. These two models are not in conflict.
