Original · Unique Research · 2026-04-14
Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay and full panel, including all named speaking turns. Revenue, growth, user, and market figures are source or speaker claims, not independently audited findings. Company, personal and product names are retained in their original English or transliterated forms where official names are unverified. The "IQ tax" (intelligence tax) and "Age of Discovery" (Age of Discovery) formulations retain the source's figurative rhetoric.
Unique Awards
Going Global in 2026: Companies Surviving on "Chatbot AI" Are Dying—How Did These Enterprises Double and Surge?
When giants start building applications themselves, what AI startups truly compete on is no longer "can you connect to a model"—but whether you can help users finish their work and genuinely earn back their money and time.
"
If you cannot help users finish their work andby the way get a good result, they will immediately judge your product as an 'intelligence tax' and decisively abandon it.
At the beginning of last year, if your AI product could answer questions well without hallucinating, users were already very satisfied.
But by this year, if you cannot help users finish their work andby the way get a good result, they will immediately judge your product as an "intelligence tax" and decisively abandon it.
This is a phenomenon frequently mentioned by several top AI going-global entrepreneurs at the recent "Unique Awards · Hangzhou AI WEEK" globalization growth roundtable.
When giants push the parameters of foundation models to outrageous levels, AI application-layer entrepreneurs are facing a hellish 2026: giants personally entering the application layer, the technology paradigm changing every two weeks, and user appetites completely inflated.
In this stage that can be called a "meat grinder," some wail that winter is coming, while others are experiencing an extremely crazy explosion.
From "Toy" to "Worker": The Cruel First Year of Agent-Native
For a long time in the past, everyone's understanding of AIremained at "advanced toy" or "knowledgeplug-in." What enterprises liked to do most was build a RAG (Retrieval-Augmented Generation) plus a Chatbot, then declare they were All in AI.
But in the view of Chen Menglin, marketing head of Zilliz (a global leader in vector databases), the wind direction has completely changed.
As a provider of underlying infrastructure, Zilliz's perception is the keenest—over the past 15 months, their ARR (Annual Recurring Revenue) has achieved an astonishing 7x growth, and on the basis of tens of millions of USD, the past three months have actually still been doubling.
"Because many European and American customers have truly moved from 'experimental testing' in peripheral businesses to 'production-grade deployment.'"
Application scenarios have crossed beyond simple Chatbots, comprehensively shifting toward Agents. Because Agents require ultra-long context, autonomous decision-making, active memory, and rapid information retrieval, this imposes extremely demanding requirements on vector databases.
Wang Yuan, founder of remio, directly set the tone for 2026: "This is the first year of the Agent-native application ecosystem."
In the future, various vertical applications will grow on Agent-native operating systems, just like APPs after the birth of the iPhone. "Previously users bought many APPs; in the future, users may just buy Tokens to 'raise their own OpenClaw,' letting the Agent help them do work," added Will, co-founder of Agnes AI.
This also means that if your product can still only do "Q&A," in this era you cannot even touch the passing line.
Giants Enter to Strangle—How Do Small Teams Survive?
What makes AI entrepreneurs most anxious in 2026 is that large language model vendors are no longer satisfied with selling APIs. They are beginning to reach into the application layer, attempting to take over all scenarios.
Facing behemoths like Google and Meta that have both money and manpower, these four companies have not only survived but thrived. How did they do it?
The first move is to go "heavier" than the giants.
Xiaoshu, marketing head of Kuse.ai (an AI productivity tool), pointed out the key to breaking through: general Agent capabilities are similar for everyone; the moat lies in extremely deep industry know-how.
They have deeply penetrated vertical fields such as education and insurance in Europe, Hong Kong, and Taiwan. "Although giants can use Skills to package, the depth of vertical domain knowledge requires extremely long exploration—this is not something a giant can break through with a general model."
The second move is to take a different path, going where giants are unwilling to go.
Giants desperately want to pull all user data to the cloud. remio does the opposite, turning itself into a native application that grows directly on the user's "primary work computer."
To truly obtain the user's primary work context, you must be in the local environment. "We have spent a lot of effort solving the parsing and extraction of local information, and currently giants have not entered to compete," Wang Yuan admitted. You need a bit of wisdom to evade the giants, but you must also maintain the excitement of challenging them in the future.
Chen Menglin went even further with a blunt and bold statement: "In 2026, there are no so-called AI giants yet."
For Google and Meta, in the AI field they are still startups. Moreover, giants tend to bet on multiple horses and run internal horse races. What truly competes with you in a specificsegmentedsector may also be a temporarily assembled small team. "In comparison, we have been deeply engaged in the vector database industry for eight or nine years—this kind of focus is something giants cannot provide."
Going Global to Make Money: You Must Know How to "Translate" Your Product
When a product has crossed the life-and-death line, how do you find users in the extremely fragmented global market and make them pay? Those who have achieved results have given very practical "money-making" advice.
First, speak the human language users understand.
Whether in the US, Europe, Japan, Korea, or Southeast Asia and Latin America, the biggest anxiety point for everyone at this stage is the same—making money.
Xiaoshu's experience is: don't tell users how great your large model is or how fast generation is. "You have to tell them how to use AI to build a one-person company for a side hustle. In every workflow, how do you help them build a website, help them do data analysis." When you translate functionality into a "money-making tool," user conversion rates are surprisingly high.
Second, trust is the only prerequisite for paying.
In the infrastructure field, trust is more expensive than anything. Zilliz won huge orders from North American giants relying on fully open-source code with 43,000 stars on GitHub, strict compliance certifications such as SOC 2, and real-person support in localized offices in the Bay Area, New York, and London.
For C-end tools, Wang Yuan's strategy is more direct to the pain point: "First achieve a commercial closed loop with a payment rate above 5%, then talk about scaled growth." Compared to buying traffic and advertising, letting bosses or HR within an organization use it well and thereby radiate to the entire company is the lowest-cost customer acquisition method under the B2B2C model.
The 2026 AI market is no longer thegrassroots era where you could raise money with a PPT or monetize by wrapping GPT.
The bubble is being squeezed out. Those who can truly eat the dividends of this Age of Discovery are the doers who dare to bite the hard bone, deeplyrooted vertical scenarios, and ultimately help users earn back their time or money.
As Will summarized in a sigh: "Every era has its heroes; the track has just begun."
More Conversation Details
Unique Awards · Hangzhou AI WEEK Trends Roundtable Panel
"Globalization: 2026 Global Growth Trends and Strategic Choices"
Guests:
Kuse.ai — Marketing Head — Xiaoshu
Agnes AI — Co-founder — Will
remio — Founder & CEO — Wang Yuan
Zilliz — Marketing Head — Chen Menglin
Moderator: EPIC Connector — Partner — Lois
Lois: It is a great honor to invite all the teachers! Today, whether it is Kuse.ai, Agnes AI, remio, or Zilliz present here, everyone is doing global entrepreneurship. Actually, we would also like to chat with everyone today about the macro trend observations of globalization seen from the beginning of the year, everyone's strategic choices, growth paths, as well as the changes that have occurred in various aspects this year and views on highlight events. To help everyone establish a new framework, the first question would like to combine the AI field your company is in. Could everyone use one keyword to describe the characteristics of global AI growth or users in 2026, and use one sentence to summarize your company's unique position in the 2026 global AI landscape? Let's start with Xiaoshu from Kuse.ai.
Xiaoshu: Hello everyone, I am Xiaoshu, marketing head of Kuse.ai. I think this is a very good question. The biggest characteristic should be "divergence." Compared to 2025, in 2026 we see more different perspectives. From the perspective of large language model vendors, everyone is working very hard to push the boundaries of capability, while there are also many new open-source projects, including Agent technology frameworks constantly emerging. From the perspective of startups or AI applications, they indeed face a more challenging situation. Because giants are entering, not only doing underlying technology frameworks but also doing more practical work related to application scenarios. So startups face greater divergence challenges in 2026 and need to find their own track and real value. Briefly introducing Kuse.ai, it is an AI productivity tool—everyone can understand it as an AI workspace. Just like in a real office, you can spread out folders, upload files, and interact with AI based on these files. At the same time, we have also introduced a new product, Junior, positioned as an AI employee for small and medium-sized enterprises.
Will: I am Will from Agnes AI. Our company mainlytargeting overseas users with an All-in-one APP that integrates efficiency tools, multimodal content generation, AI character companionship, and other capabilities. Starting from March this year, we began splitting our APP matrix: launched Agnes Echo focused on companionship attributes, and Agnes Pixa focused on image and video models. In April, we will also launch Agnes Vault focused on website building and Coding. In 2025, we continuously accumulated model capabilities and expanded our team. Answering the moderator's question, my feeling is "paradigm shift." As technology and frameworks change, user behavior is also shifting: in the future, more people will call models billed by Token through "raising OpenClaw," rather than subscribing toa large number of applications. This may be a paradigm shift in 2026. Pure beginners may not be suitable for directly "raising OpenClaw," so we have launched the one-click deployment capability of Agnes Cloud,convenient non-programming users to quickly experience and reduce costs. In terms of company positioning, we are transforming from an application company with model training capabilities to a model company with application capabilities.
Wang Yuan: Hello everyone, I am Wang Yuan, founder of remio. Briefly speaking about the remio product, currently it is an office-focused product that can seamlessly collect and aggregate context and various data, and supports knowledge base Q&A. Next week we will release a new product, which is an Agent-native application operating system and application marketplace. 2025 was called the first year of Agent; I believe 2026 is "the first year of the Agent-native application ecosystem." In Silicon Valley, many vertical applications have emerged, and the form will evolve into native applications running on Agent-native operating systems, just like the series of mobile APPs produced after the birth of the iPhone back then. So I believe this year is the node for application explosion.
Chen Menglin: Hello everyone, I am Chen Menglin, marketing head of Zilliz. Zilliz is a global vector database leader. My feeling may be different from the previousseveral. For Zilliz, this is very directly "a year of massive explosion." Over the past 15 months, our ARR has achieved 7x growth; on the basis of ARR reaching tens of millions of USD, the past three months have still achieved doubling growth. Regarding Zilliz's position in the Agentic AI landscape, at last week's GTC conference, Jensen Huang展示 the ecosystem landscape of unstructured data, in which Zilliz and Milvus were highlighted as very important links. This proves that Zilliz itself is an indispensable link in the data infrastructure of the Agentic AI era.
Lois: Zilliz was also nominated by Jensen Huang before. While achieving such significant growth last year, when communicating with overseas customers, have you seen some interesting changes in customer needs or technology stacks?
Chen Menglin: I think this question is very interesting. To summarize simply, it is that global customers are all moving from experimental practice to truly production-grade deployment. One or two years ago, everyone mostly did experimental testing in peripheral businesses; this year, more and more enterprises are moving their most core business and data flows to vector databases. Therefore, everyone values security and compliance requirements more, enterprise-grade high availability, and real-time implementation support services. At the application level, previously everyone defined vector databases as RAG's knowledgeplug-in or long-term memory; today market applications have basically shifted from chatbot to Agent. In the Agent context, long-context capability, autonomous decision-making, active memory, rapid information retrieval, and security compliance impose higher requirements on the vector database itself.
Lois: Indeed, from the infrastructure perspective, security, stability, and efficiency are what everyone pays more attention to. From the application and user perspective, I would like to ask Teacher Wang Yuan from remio: have you seen some changes in user habits?
Wang Yuan: A clear change is that in the first half of last year, users were relatively satisfied if AI software could answer questions well; by the second half of last year, especially after OpenClaw's global influence expanded, users expressed that if you cannot help me do work, they would be very dissatisfied. At this stage, whether doing vertical or general products, the bottom line is to let users放心 hand over their work to you and get good results.
Lois: Requirements are getting higher and higher. I would also like to ask Will to share—Agnes AI has very wide overseas user coverage. Are there any trends in this regard?
Will: 50%–60% of our users are in Southeast Asia, 30% in Latin America and the Middle East, and the remaining 20% in the US, Europe, Japan, and Korea. I very much agree with Teacher Wang's view that Trust between product and user is very important. Building Trust is based on two points: first, the completion rate of complex tasks, and second, whether the cost under the same conditions is accepted by the market. In early 2025, general Agents were popular; mid-year was various Artifact generation; from the end of the year to early 2026 was multimodal and world models. After OpenClaw came out, it brought a paradigm change: users can more flexibly build their own Agents, handle longer workflow tasks, and meet personalized needs. We hope that by providing APIs, users can experience the process of "raising OpenClaw" themselves with better performance and cost-effectiveness.
Lois: Indeed, the paradigm change brought by OpenClaw is quite large—it not only raised expectations but also prompted everyone to adjust their products. I would like to ask Xiaoshu from Kuse.ai: you recently completed a new version refactoring. What are your next steps to satisfy users?
Xiaoshu: I very much agree with the judgments of the previous three. From a marketing perspective, for example: in 2025, doing social media promotion only needed to explain what the product could help users accomplish; in 2026, AI penetration is very high, and you must give users a reason for "why use you instead of others." On the product side, users are no longer satisfied with a simple good chat experience; they expect more Agentic experiences and to弥补 experience breakpoints at every generation node. For example, ordinary people cannot handle APIs, but in OpenClaw, if they can connect to external databases, capabilities will grow rapidly. We hope to internalize these capabilities into the product, helping non-technical people弥补 experience breakpoints, which is a very large value gain.
Lois: Very sharp observation! Actually, everyone is thinking about a question: now top giants have model capabilities, and every update may颠覆 the industry. How do you view the competition brought by giants entering the application space? Starting with Xiaoshu.
Xiaoshu: Giants have larger human and financial capital, and their release of new features is indeed leading. At this time, following is not a derogatory term, because AI development is still in a very early stage. As a startup, on the basis of doing well in general Agentic capabilities, we will do more vertical innovation. We have penetrated relatively early in education and finance (especially the insurance field), and have in-depth cooperation with schools and insurance institutions in Europe, Hong Kong, Taiwan, and other places. Vertical needs are very different from general ones and require extremely deep industry know-how. Although giants can package through Skills or Connectors, the depth of vertical domain knowledge still requires a lot of time to explore.
Lois: The more vertical, the more it can become a moat. Then how does Agnes AI find development space and retain users?
Will: The AI application layer is still in a relatively early stage, the track has just begun, and every era has its heroes. There is differentiation in market choice, which must be considered based on the user needs, pain points, and willingness to pay in the selected market. Small teams have the opportunity to grow quickly and compete with giants. Giants also have dilemmas—when their scale is large enough, they consider more than just a single AI product, so small companies have more opportunities.
Lois: Teacher Wang Yuan from remio, in the knowledge management direction, how do you find space for continuous development?
Wang Yuan: As a startup, there is no need to position yourself in a place where you absolutely believe giants will not come. Our main competitors are large model companies, office collaboration companies, and internet giants. The common point of these companies is that they desperately want user data. They cannot accept data only being saved on the user's local personal computer and not going to the cloud. We are a producttargeting individuals, directly growing on the user's primary work computer. For example, OpenClaw has also proven that to truly get close to work context, you must be in the local environment. We have spent a lot of effort solving the parsing and extraction of various local information, doing a good job in context engineering. Currently giants have not entered to compete, but in the future large models will definitely also move toward localdevice, bringing competitive pressure. Entrepreneurs must both evade giants and have the excitement of challenging them.
Lois: Thank you, Teacher Wang Yuan, for your advice. For Zilliz, because we have offices globally and it is a To B model, how do you view the challenge and key solutions of building trust with customers in different regions?
Chen Menglin: Let me first补充 on the giant question: I have a bold statement—in 2026, there are no so-called AI giants yet. For Google and Meta, in AI they are still startups. The currently popular applications have all developed from startups, and 2026 is a good opportunity for everyone. Although giants have many resources, they bet on multiple horses, and specifically in a certainsector it is still small teams competing. Compared to the temporarily assembled teams of giants, our team has been deeply engaged in the industry for eight or nine years. We were also rated by Forrester as a global vector database leader, defeating traditional vendors such as AWS and Oracle. Regarding building a globally trusted brand, there are three points: first is open source—Milvus currently has over 43,000 stars on GitHub, and we donated it to the Linux Foundation, with all code transparent. Second is security and compliance—we have compliance certifications such as SOC 2 Type 2 and GDPR. Third is localized operations—we have offices and localized architects in the Bay Area, Seattle, New York, London, Tokyo, and Singapore, providing face-to-face support.
Lois: Very much agree—deep engagement in vertical industries and localized trust are great moats. Finally, time is limited. Can everyone share your exclusive tips or new玩法 on growth? Xiaoshu first.
Xiaoshu: Growth is about translating the product into language users understand. No matter which country's audience, the biggest anxiety point at this stage is "making money and entrepreneurship." Tell them how to use AI to build a one-person company for a side hustle, helping them build websites, do Branding, or data analysis in every workflow—users are more receptive. In terms of channels, we mainly rely on UGC Creator cooperation, holding daily standups with them, analyzing viral logic, and globally replicating and promoting. In the future, perhaps marketing content can be automatically generated directly through AI Agents.
Will: Since launching in July 2025, we currently have over 7 million registered users. Our vision is AI Parity, AI Inclusion, and AI Neutrality. Wetargeting users who desire AI technology but lack access channels. To reduce costs, we chose the route of self-developed training models. Our Agnes Cloud series models perform well on various benchmarks. The next core of growth is: letting users access our models through mobile and webdevice, and letting users call the models through third-party API routing platforms.
Wang Yuan: We first completed the verification of the commercial closed loop (payment rate over 5%) before proceeding with scaled growth. In terms of customer acquisition, I believe KOL marketing currently has relatively low cost-effectiveness; SEO can bring持续 natural traffic returns; advertising is still the most scientific method. With the help of various Agents, we can more efficiently adjust advertising strategies. However, ultimately we still need to desperately light up the product's "tech tree," relying on stunning breakout effects forspread. Additionally, we must dig deep into internal organizationalspread (B2B2C model)—many times, when the boss or HR uses it well, it will连带 radiate to the entire company for conversion.
Chen Menglin: Let me share the breakdown of the Funnel theory. At the top level is the open-source ecosystem—on the basis of 43,000 GitHub stars, we cover over 100,000 enterprise deployments, which is the largest pool for business opportunity conversion. The middle layer is PLG (Product-Led Growth)—Zilliz Cloud adopts a Pay-as-you-go pricing system and a极简 onboarding process, letting users seamlessly convert to paying customers. The bottom layertargeting large-budget top customers (such as DoorDash in North America), relying on deep localized teams and services (GTM strategy) to conquer. This combination of punches has brought Zilliz very healthy commercial conversion growth.