"Over the past three years, our thinking has been stagnant, but after using this system, growth has been very rapid, achieving 40% to 50% growth, and by the end of the year, we estimate we can achieve 100% growth in a single month. The strategy hasn't changed; it's just ensuring that everything we do is fully focused on achieving strategic goals." Wang Zhiguo, CEO of Coocaa Technology, used a set of data to intuitively illustrate the transformation brought by the "AI-native enterprise operating system."
In May this year, Coocaa officially launched Happy Work AIOS Lite MVP, and "silicon-based management" along with the "four-agent architecture" drew industry attention. However, there is still significant confusion in the market about the concept of "enterprise AI operating systems"—often equating it with AI collaboration plug-ins or general Agent tools, failing to distinguish between "bolt-on + AI" and the fundamental difference of underlying AI-native reconstruction.
With three questions—"how will the track landscape evolve," "what is Coocaa AIOS's moat," and "data ownership of personal intelligent agents"—a reporter from China Economic Net conducted an exclusive interview with Wang Zhiguo.
ERP is irreversible, but the school of thought lies in "partial" versus "systemic"
When asked about the future landscape of the AI-native enterprise operating system track, Wang Zhiguo's judgment was straightforward: "It's definitely irreversible."
He observed that traditional enterprise management software companies are all agentifying their capabilities. "Our real productivity now lies in turning capabilities into Agents for others to call, thereby enhancing the competitiveness of enterprises in the AI era." But he pointed out that the only difference in this process is whether the Agents you build solve partial problems or systemic problems.
Wang Zhiguo used the development history of mobile phones as an example: "Before smartphones came out, we had phones with touchscreens, but they didn't take off. The simple reason was that the phones of that era had no ecosystem; all development was implemented by the manufacturer, and the manufacturer couldn't solve the different situations of different enterprises."
He further explained the difference between "partial" and "systemic" using his own experience: "Most of my TV users are middle-aged and elderly, but the team is all post-90s. Expecting them to understand the preferences of middle-aged and elderly people and create a TV they love is impossible. But AI can, because AI understands middle-aged and elderly people."
"Happy Life hands over all TV operations to AI, and all enterprise management to AI," Wang Zhiguo said. "I am a post-80s manager, and I can no longer manage the post-90s and post-00s in the company. I can't talk to them about sentiment; they care a lot about actual gains, and ideals don't work either."
He analogized this intergenerational communication dilemma to the relationship between parents and children: "We love our children, but do children feel that their parents love them? Not necessarily. Our love is misunderstood. AI can help resolve this difference in our interactions."
"AI-native as the underlying operating system essentially solves the original differences in communication between people and addresses the problem of the 'four gaps' in corporate strategy," Wang Zhiguo emphasized. "The AI that traditional enterprises do is partial AI, not solving fundamental capabilities. If you start from an AI-native perspective, your skills will surely be useful."
Differentiation barriers go beyond architecture and industry advantages; the key is "external thinking"
Coocaa was born from the home appliance giant Skyworth—does this constitute a unique advantage? Facing the reporter's question "Is the barrier more about product architecture or industry advantages?" Wang Zhiguo's answer was unexpected.
"I would rather experiment externally," he said. "The company itself is an experimental subject because I need to solve my management problems. But external experiments make it easier to break away from traditional thinking."
He revealed that Coocaa found a completely different track—the fast-moving consumer goods industry—to experiment. "Just to verify whether this method works. The result proved that the effect was even better than in our own company."
He used a detail to reveal the difference between internal and external thinking: "Our HR and finance departments spent a long time helping companies sort out business models and value chains, but even now, they haven't sorted it out as completely as they did for the external FMCG industry."
Wang Zhiguo frankly admitted that industrialization is not for making money, "but rather I hope the team uses external thinking to truly realize the value of things, rather than internal self-congratulatory products."
In his view, the real barrier is not industry background, but "the courage to give up experience and inertia and let AI do systematic sorting." He said: "The logic of thousands of industries is different, but AI can exactly help us break through inertia. If you don't understand the industry, you won't come with bias, and the system sorted by AI is much higher than manual sorting."
Data attribution cannot be "shortchanged," how to part ways amicably with personal abilities and job assets
Personal intelligent agents both accumulate employee knowledge and can be reused across organizations. How can enterprises avoid capability loss? How can employees prevent their digital assets from being "taken for free"? This is the third question the reporter raised, and also one of the most sensitive topics in the organizational transformation of the AI era.
Wang Zhiguo admitted that in the previous version, employees and positions were not separated. "All responsibilities were on the position. When the employee left, they left without anything of their own."
"Someone told me honestly: if the system makes employees contribute everything and then suddenly optimizes them out, and the agent does the work itself, who would be willing to contribute?" Wang Zhiguo said. "Later I thought, this is the fundamental problem AI must solve."
The new architecture design embodies a "two-way protection" logic. He explained that personal intelligent agents are registered and trained on the public network, not controlled by any enterprise. When an employee works for a certain enterprise, they establish a clone bound to the enterprise's position, inheriting the personal agent's capabilities and then doing job tuning.
"All things effective for the position will settle into the position. When the employee leaves, the capability at least stays at the level before leaving. Meanwhile, the feedback during the clone process—including thinking ability, judgment ability, and expression style—will settle into the personal intelligent agent."
When decoupling, the clone strips the enterprise information, and the enterprise strips the employee's capabilities. "There's a complex mechanism inside; we designed a task system where both parties' capabilities interact in the task system, and then the task agent settles the capabilities into their respective sides."
Wang Zhiguo said this mechanism is designed to ensure both personal capabilities and job capabilities are reasonably settled, so neither party "loses out."
Racing ahead with first-mover advantage, the true moat of AI-native systems lies in "delegating power"
Regarding the deployment cost and timeline that enterprises are concerned about, Wang Zhiguo also gave a clear path: the first stage is deployment, training personal clones, and aligning strategic information, taking about three months; the second stage connects systems such as ERP and HR to form input-output analysis and result closure, requiring another three months; then follows a parallel period of about six months of "AI simulation assessment" and manual evaluation to verify fairness before silicon-based management truly lands.
Regarding costs, he frankly said "AI costs are quite high." Currently, Coocaa internally consumes about 7 million Token fees per month. After running for a period, they will do knowledge distillation and eventually compress to maybe two to three billion parameters, significantly reducing operating costs.
As for the return on investment that enterprises care about most, Wang Zhiguo is full of confidence: "Just the first step—efficiency improvement in the first three months—will allow enterprises to double their investment in the system."
When asked about the relationship between open ecosystem and patent moats, Wang Zhiguo was quite candid. He said the team originally sorted out about 50 patents, but later decided to cancel most of them. "Good things are not afraid of others copying; things that are afraid of copying are not good either."
"If Tencent were sitting with me today and we both talked about the same concept, who would you choose? You would definitely choose Tencent," Wang Zhiguo said. "So I must first race ahead with a first-mover advantage."
In his view, the true moat of AI-native systems is not technology, not architecture, but management philosophy and deep understanding of organizational transformation—and the courage to "delegate your power and break all your boundaries."
"If you don't dare, this system cannot be built," Wang Zhiguo asserted. (China Economic Net, Reporter Zhao Han)
