Original · Unique Research · 2026-04-17
Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, repeated examples and full panel, including all named speaking turns and their continuation paragraphs. Revenue, user, performance, market and product figures are source or speaker claims, not independently audited findings. Company, personal and product names are transliterated where official English forms remain unverified. OpenClaw is referred to as an open-source Agent framework in the source; "Personal crayfish" (little lobster) is a colloquial nickname used by speakers. The source is dated April 17, 2026. Statements about product rankings, market sizes and future projections are attributed to speakers, not independently established here.
Unique Awards
When Software Is No Longer Designed for Humans: The Agent-Native Product Revolution
The real change is not just that software has more AI in it—it is that "who uses software" is being entirely rewritten.
"
In the future, software will likely have to serve two completely different kinds of users at the same time.
One is humans. The other is Agents.
The moderator asked the entrepreneurs on stage: if in three or six months, among the traffic entering your products, there are more Agents than humans, what will you do? Indeed, since OpenClaw came out, some companies are already considering which "lobster" to lay off.
Over the past few years, when we talked about AI products, the default premise was always that "humans" were using the software. It was nothing more than humans clicking buttons, humans writing prompts, humans adjusting parameters—humans sitting in front of a screen, letting AI do a little more work for them. But what was truly interesting about this roundtable was that the guests were all,in unison地, acknowledging something else: going forward, the direct user of much software may no longer be primarily humans, but Agents.
This means a very fundamental question has been dug up.
Who, exactly, should software be designed for?
If the software revolutions of the past went from command lines to graphical interfaces, from expert software to mass-market software, then this Agent round may be an even more thorough deflection: software is slowly shifting from "being operated by humans" to "being called by Agents."
Not Helping You Think—Doing It for You
Wang Jiancong, co-founder of AhaCreator, gave a very typical entrepreneur's answer. He said what they have been doing is actually not some trendy concept, but execution. When helping clients do influencer marketing overseas, they automate outreach, automate bargaining, automatically send briefs, and handle the dirty, tiring work that used to be piled up by operations staff, salespeople, ad buyers and interns. In his words, even before there was the concept of OpenClaw, they had always been emphasizing "execution."
Buried in this passage is actually a very important change.
In the past, many AI products liked to package themselves as "assistants," as "co-pilots," as "inspiration tools." To put it more directly, most of them stillremained at helping human users "use software more easily." But what Wang Jiancong was talking about was not that. He was talking about swallowing an entire segment of actions that originally needed humans to complete into the system as much as possible. This is actually already an early prototype of Agent Native. Not helping you think—doing it for you.
Even more noteworthy was a very specific on-site observation he gave: some users have already started using only OpenClaw to operate their platform.
The weight of this sentence is actually heavier than any industry judgment.
Because it shows that the so-called "Agent becoming the user" is no longer just a future trend being discussed on stage, but something sporadic yet real happening in the present tense. Even users' perceptions are changing. Last year, advertisers were still saying that if influencers used AI to make content, they would definitely not collaborate; today, some people already feel that AI-generated content may be better than what humans can do.
Changes in many industries start with capability changes, thentransmission to perception changes, and finally rewrite the commercial structure. Looking at it now, the two links of influencer marketing and content production have already started entering this chain.
Not Making a Tool—Making the Whole Closed Loop
Of course, this change is not monolithic.
Wang Ming's attitude was more radical and more provocative. His K2 Lab does another kind of thing: helping ordinary users, especially everyday people and C-end users, amplify their originally weak monetization ability. Their first publicly disclosed data was that, after one month of exploration in the US, they helped 30 people earn US$300,000.
This figure is certainly eye-catching, but what is really worth noting is not the number itself, but the method he described afterward.
In his definition, for an AI product to truly break through, it must either provide a "ten-thousand-fold efficiency" tool, or turn a certain scenario into an end-to-end closed loop so that humans barely need to do any work. He talked about how when he used to travel, he had to send many emails to travel agencies and guides, and make complex spreadsheets; when he wanted to share his experience with others, he couldn't make structured content, couldn't write scripts or edit videos. So they strung together script creation, automatic publishing and automatic analysis, letting users make only a small number of choices, with the rest of the processmanaged as much as possible.
To put it plainly, he is not selling a video tool, and not just selling a content tool—he is selling an entire chain "from monetization intent to result delivery."
So when Zhu He pressed: if you let OpenClaw open a browser now, can it do all your work for the user?
Wang Ming's answer was very direct: No. Because we have already built something more advanced than OpenClaw. Today OpenClaw is still a toy.
Thisevaluation will certainly be controversial, but the controversy itself precisely shows that the core of the problem has surfaced.
One type of entrepreneur judges that general-purpose Agents are important, but they can only do scheduling, coordination and entry points; what can truly deliver results is still a deeply polished system in a vertical scenario. Another type of entrepreneur judges that since general-purpose Agents will become the new super entry point, products should be designed for Agents from the very beginning, becoming a callable part of their ecosystem.
From Day 1, Reserve a Place for Agents
Xu Anbang, founder of LoovaAI, clearly leans toward the latter.
He said very frankly that their company is relatively new, founded right when OpenClaw appeared, so from Day 1 they have designed their product according to two systems: one for humans, one for Agents. At the current stage, of course, humans still use it more, and the interface is indispensable; but at the same time, they will make their product into a Skill,integrate OpenClaw's video ecosystem position. As it develops, human usage will decrease and Agent usage will increase.
Because it represents a very "new world" product philosophy: no longer treating Agents as anaffiliated channel, but treating Agents as first-class users. From this perspective, future software design may not be as simple as adding one more set of APIs or one more developer document, but requires re-splitting capability modules at theunderlying, rethinking what is suitable for human operation, what is suitable for Agent scheduling, and what must exist in some form of "space."
The "canvas" Xu Anbang mentioned is a typical example.
He said that in video creation, Skills currently can only solve part of model calling, and cannot solve the canvas problem. Zhu Heconveniently translated this viewpoint very vividly: if MD documents are text space, then the canvas is multimodal space. Skills must act on top of this space.
This has actually pointed to a key layering of many future Agent products.
One layer is callable capabilities. Generation, analysis, retrieval, orchestration—these things can all be interfaced, Skill-ified and modularized. The other layer is the work interface that cannot be easily abstracted away. For example, the timeline in video, the canvas in design, the status panel in complex workflows. They are not simple collections of buttons, but part of cognitive space.
Whoever can build this space will not just be a called capability provider, but may become part of the new entry point.
Beyond the Entry Point, the Real Difficulty Is Holding the Closed Loop
Returning to this conversation, a particularly interesting point is that although the guests did not completely agree on their judgments of OpenClaw, they alldefault one thing: the traffic structure will change.
This is already very different from the intuition of the previous generation of internet entrepreneurship.
In the past, when we made products, we fought over where humans came from. Search, app stores, social distribution, paid traffic, KOLs, SEO—all revolved around "bringing humans in." Now the question is slowly becoming: besides bringing humans in, can you be found, understood, called, reused and持续 used by Agents?
Xu Anbang's answer was to embrace SEO, embrace the most core Agent entry points, become part of them, rather than building a new entry point yourself.
Wang Jiancong's answer was more realistic. He said the product will not change much, because what they always do is optimize influencer matching and price matching. If OpenClaw comes to call, let it call; the dirty outreach work in the middle is still done by them. GTM will not change either, because OpenClaw still has an owner, and the ultimate object of their Marketing is still the owner of OpenClaw. Whether the decision-maker is a CMO or an intern, the person making decisions in the future will still be a human.
Frankly, I really like this answer, because it was not carried away by the "entry point illusion."
When many people talk about Agents, they tend to think of all problems as "who occupies the entry point." But in reality, no matter how entry points change, what is hardest to replace in the business world is often those capabilities that are truly hard to do, hard to accumulate and hard to migrate. For AhaCreator, this capability may not be the interface or the slogan, but outreach capability, influencer pricing models, and the data and experience continuouslyaccumulated behind them. Wang Jiancong gave a very specific example: influencer pricing does not only look at static indicators like country and follower count, but also judges follower profiles and geographic location, and combines dynamic market conditions and historical collaboration data to give a floor-price model.
This is very much like the core capability of ride-hailing platforms. On the surface you see "call a car," but what truly supports the experience and profit is the complex supply-demand matching and dynamic pricing underneath.
And so, when Zhu He asked, if a client only wants your bidding capability and splits you into several API calls, do you support that?
Wang Jiancong's answer was: No, I don't support it.
He put it very vividly: This is like ride-hailing. My service is to bring the car to you. You can't just use my Agent to check a price but not take the car.
This sentence actually directly hits a new难题 that many products in the Agent era will face: when all capabilities can be split, orchestrated and aggregated, are you willing to become a pluggable part, or do you want to find a way to hold the complete service closed loop?
The former means easier integration into ecosystems and easier access to new traffic; the latter means stronger pricing power and more complete value capture. There is no absolute right or wrong, but different choices will lead companies to completely different endgames.
From "Humans Opening Software" to "Which Agent Gets Called First"
Wang Ming's judgment went one step further.
He said that in the future, everyone may have their own Personal "little lobster," that is, their own OpenClaw. All centralized entry points may disappear, meaning they want to become the Agent OS for a certain group. If OpenClaw solves the long-term memory problem,superimposed the rapid iteration of the open-source ecosystem, it may become the most important Personal AI open-source framework of the next era.
This sounds grand, but it is not empty.
Because the so-called "disappearance of centralized entry points" essentially means that platform distribution logic may be rewritten again. In the PC era, entry points were concentrated in portals and search; in the mobile era, entry points shifted to super apps and app stores; in the AI era, if everyone has a Personal Agent with long-term memory, the ability to call tools, understand preferences and manage tasks, then many traditional entry points will indeed be weakened. Users no longer personally open a dozen apps to complete tasks, but hand their goals to their own Agent, and the latter handles cross-platform scheduling.
If this day truly comes, the most important question for the software industry will no longer be "whose app did the user open today," but "whose service did the user's Agent prioritize calling today."
Once this perspective shifts, many product decisions that seem minor today will have their significance amplified.
For example, have you left an interface for Agents?
For example, can your product capabilities still stand after being split apart, or must they exist as a closed loop?
For example, is your interface just an operation console for humans to look at, or a shared space for Agent-human collaboration?
And for example, is your moat ultimately prompts, workflows, data, model scheduling, canvas, or some irreplaceable result-delivery capability?
The Real Change Is That "Who Uses Software" Has Been Rewritten
Ultimately, what I really cared about in this roundtable was not the specific name OpenClaw from the entrepreneurs' mouths, but that they had all started reorganizing their businesses around a common premise: humans are slowly shifting from "direct operators" to "goal-setters" and "final decision-makers."
This will bring a very realistic consequence.
In the future, software will likely have to serve two completely different kinds of users at the same time.
One is humans.
They need to understand, confirm, correct, apply aesthetics and make final decisions.
The other is Agents.
They need to call, read, execute, coordinate and provide continuous feedback.
Much software that seems fairly usable today may not be enough by then. Because they are suitable for humans to click, not for Agents to call; suitable for humans to look at, not for Agents to understand; suitable for displaying features, not for delivering results.
In other words, the future is not just "adding AI into products," but "the product structure itself must be redone for AI."
This also explains why, in this roundtable, what everyone repeatedly mentioned was not a single model capability, but words like execution, closed loop, Skill, canvas, entry point, pricing, ecosystem position, Agent OS. Because the real competition has already moved down one layer from "who can connect to a large model." What is being competed on now is who can truly turn large models into a stable, reusable, collaborative and profitable production relationship.
This may sound a bit grand, but returning to the scene, everyone has actually already given their own answer.
Wang Jiancong's answer is: first thoroughly do the dirtiest, most tiring and hardest-to-scale executionaspect; it doesn't matter who schedules me.
Xu Anbang's answer is: reserve a place for Agents from Day 1, design both human and Agent systems together, and compete for the next-generation video creation entry point.
Wang Ming's answer is: don't deify general-purpose Agents; what truly makes money is still end-to-end system organization capability in vertical scenarios. Whoever can捏 together Agents, Skills, workflows and commercial closed loops has the资格 to be the Agent OS for a certain group.
And Zhu He's closing summary actually also hit the theme.
He said many people may think he is overly optimistic about Agent traffic exceeding human traffic. But the internet has hundreds of millions of users, mobile internet has billions, and in the Agent era, it may be on the order of tens or even hundreds of billions.
You can disagree with this numerical judgment, but it is hard to ignore the reminder behind it: today's product teams, entrepreneurs and developers should probably all start seriously thinking about one thing.
Can your product let OpenClaw use it for you?
If not, then what you are facing may still be the previous generation of software logic.
If yes, don't rush to celebrate, because the next question comes immediately: are you just a called link in the chain, or that irreplaceable key node?
This is what is truly cruel—and truly fascinating—about the Agent-native product revolution.
It is not about making software a little smarter, but about entirely rewriting "who uses software."
And all those who move forward will sooner or later have to answer this question.
More Conversation Details
Unique Awards · Hangzhou AI WEEK Trends Roundtable Panel
"When Software Is No Longer Designed for Humans: The Agent-Native Product Revolution"
Guests:
AhaCreator — Co-founder — Wang Jiancong (Wels)
LoovaAI — Founder — Xu Anbang
K2 Lab — Founder — Wang Ming
Moderator: Yeahmobi — AI Head — Zhu He
Zhu He: AhaCreator was super hot this time last year, because you shouted "you no longer need a marketing team, just use it, it's an All-in-one AI CMO." A whole year has passed today—have you helped clients largely achieve this? You also heard client feedback just now below.
Wang Jiancong: To answer this question seriously, we are achieving this step by step, and it's not just us achieving it—clients are achieving it too. And Teacher Wu is particularly amazing for gathering together everyone who seems to want to achieve this today. In our current pipeline, we have started having people who play with OpenClaw and use only OpenClaw to operate our platform. Last year all advertisers said if influencers used AI to make content they would definitely not collaborate; but today some people are starting to feel that AI-made content seems to have reached a state even better than what humans can do.
Hello everyone, I am Xiao Wang from AhaCreator. This is a "new" startup company, because we have been changing names constantly over the past two years and often get sued. A piece of startup advice: if you make overseas products, you must pay attention to trademark registration. What we have actually always been doing is one thing: helping everyone do influencer marketing overseas by automating outreach, automating bargaining, doing this dirty tiring work, automatically sending briefs and communicating with influencers. Actually, even before there was the concept of OpenClaw, we had always been emphasizing doing "execution."
Zhu He: Speaking of changing names, as far as I know there have been at least four or five. Wasn't it originally called Head? Why did you change it back again?
Wang Jiancong: Because I am an engineer and never thought before that a brand name could really affect how clients see us. Last night I was chatting with an American client, and he said: "If you call yourselves Aha again we'll use you; if you use your current name we won't pay." Quite outrageous. But a while ago after we changed back to Aha, we got sued again. In the US there is a Swedish advertising company called Aha! that has been around for 11 years, and slap—a lawyer's letter arrived at our American colleagues' office again. So these days we are changing names again, this time to AhaCreator, and the graphic trademarks in front have also been registered. Hope we don't have to change again this time, otherwise all the previous SEO will be wasted.
Zhu He: I think that's pretty good, it also proves your知名度 has gotten out there. I at one point even thought you had sent too much spam email and your domain got blocked, which is why you kept changing names.
Wang Jiancong: We don't use our own domain for sending emails! We have over 1 million email addresses.
Zhu He: Anbang, let me call on you too. You went to Silicon Valley, and after coming back you said others sell rough-processed Tokens—just adding a Prompt to let others generate digital humans; this time you came out wanting to make deep-processed Tokens, using Agents to empower various video fields. This new project is actually about making the next-generation video creation entry point, an All-in-one Video Agent. Your new project is only one month old, and the day before yesterday an investor told me your revenue is taking off—convenient to say what your current traffic and MRR are?
Xu Anbang: No problem, we're all among ourselves. Our project has been live for less than a little over a month, and now we should have US$100,000 in MRR. The previous project took about nearly nine months to reach US$100,000; now the new project did it in one month, which shows everyone wants to scale up now.
Zhu He: I taught him how to do influencers last time around—no wonder he ignored my WeChat requests back then, it's fine now. All the big shots are using your products. Actually, whether you are doing Agents or education or selling APIs, everyone is essentially "selling Tokens." I read an article before saying that simply selling electricity, turning it into electrolytic aluminum and selling electricity, and deep-processing electricity have efficiencies of 1 to 6 to several dozen. Anbang's kind of deep-processed Token, compared to the shallow Prompt rough-processing that everyone used to do with wrapper products, will indeed go deeper.
Xu Anbang: Exactly. Now there are different processing methods for selling Tokens: one is rough processing, adding a Prompt template for others to use; the second is becoming an Agent in a vertical scenario, directly solving problems; the third is deploying models on servers for inference acceleration.
Zhu He: Welcome our third guest, Mr. Wang—introduce yourself in one sentence.
Wang Ming: I am an explorer full of passion in the AI era. Including choosing to leave a big company to start a business, it was all because of relatively radical exploration. What our company does is help everyday people or C-end users effectively amplify their potential monetization ability, letting them directly make money. We explored in the US for one month, and this is the first time we are disclosing data: in one month, we helped 30 people earn US$300,000.
Zhu He: Pure everyday people? They might have only made US$100 a year before, that's really amazing—how did you do it?
Wang Ming: When I was at a big company, I once proposed that we should build China's largest AI venture incubator. We spent about half a year looking at over 400 projects and found that AI can already achieve PMF (Product-Market Fit) in many scenarios. Today, to make a product that breaks through very quickly, you either need a ten-thousand-fold efficiency tool (like Coding), or end-to-end in a certain scenario where humans have to do very, very little. We organically combine these AI capabilities to help users solve end-to-end problems. For example, when I used to travel, to get cheap prices I sent countless emails to travel agencies and guides, and made very complex spreadsheets. I wanted to share with others but I couldn't make structured content, couldn't write scripts or edit videos. So we helped them make up for this part.
Zhu He: So yours is helping influencers do video editing? Fully automatic?
Wang Ming: Video editing basically accounts for only a very small part. To summarize: the full closed loop of full-product script creation, automated publishing and automated analysis—basically no humans are doing the work. After a user enters our APP andmanaged their account, we analyze their follower profile and content tone. Then we use large models for product selection, match their tags, let them make a few choices, and the whole process is strung together.
Zhu He: Then here is my question. If you use OpenClaw to open a browser now, can it complete all the work for this user?
Wang Ming: I don't think so. Because essentially we have already built something more advanced than OpenClaw. Today OpenClaw is still a toy; what these big companies have launched are actually things we did back in the day. OpenClaw's value today is its open-source value and its value in calling local applications, but it is not a productivity-level thing.
Zhu He: I think your understanding is wrong. I believe OpenClaw's role in your project should be that of a project manager. It's not about letting it select products better than you can, but letting OpenClaw use a project manager's mindset to manage what this user should do.
Wang Ming: I agree, so it is a very low-end butler. Back when we built a similar foundation, we also hoped that experts in varioussegmented fields would come in and build truly awesome Skills, so that it could coordinate and schedule. Big companies have very poor strategic定力 in many of their moves.
Zhu He: What do you two think about your products? Back to Anbang—now if an OpenClaw or Claude goes in, can it use your product?
Xu Anbang: My view may be different from Wang Ming's, perhaps because our company is relatively new and was founded right when OpenClaw appeared. So from Day 1 we designed two sets: one for humans, one for Agents. At this point in time, humans definitely use it more, and the interface is indispensable, but at the same time we will definitely make it into a Skill, becoming the video ecosystem positionintegrate OpenClaw. As it develops, human usage will decrease and Agent usage will increase.
Zhu He: Then how do you face this kind of situation in the future? Suppose in three or six months, traffic from OpenClaw exceeds human traffic? Just like when I made a website with zero promotion before, 70% of the traffic all came from ChatGPT.
Xu Anbang: The first is SEO, the second is you have to embrace the most core Agent entry points now. You have to become part of them, rather than building a new entry point yourself. At the current stage, for us to do video, secondary editing and canvas (Canvas) are indispensable; Skills currently can only solve the model-calling part, and cannot solve the canvas problem.
Zhu He: Let me understand—canvas is like how MD documents are a space for text, canvas is a space for multimodality. Skills must act on top of this space.
Xu Anbang: Exactly. If you do the canvas well, in the future you will become a plugin for the Agent entry point. We want to make the next-generation video entry point.
Zhu He: Very much looking forward to you becoming the Final Cut Pro here. I just saw on Vercel's skills.sh marketplace that ranked fifth is the video editing Skill: Remotion. I think you have great potential to surpass it. OpenClaw and these AI entry points have basically completely revived Remotion. Jiancong, you just said some clients are already using Agents to fully schedule this, without human participation—how amazing is this?
Wang Jiancong: Operationally it's actually not difficult; the key is everyone's ideology. For us, whether humans use it or Agents use it is like whether a ride-hailing app logs in on a phone or a computer—in the end the driver will arrive in front of you. That day I asked a client: "Do you really trust OpenClaw to select influencers for you?" That client said: "If it doesn't select well, I can train it." He felt that poor performance at the start is experience accumulation, not loss—which left me with no reason to refuse. This should be something everyone with a foundation can play with. My mom earned US$2,000 in two weeksrely on AI, and I hope everyone can use it—that is what a product in the AI Native era looks like.
Zhu He: Then going back to the question I asked Anbang earlier, if in the future most clients use OpenClaw to use your system, what will you do in terms of product and GTM?
Wang Jiancong: The product will not change much. We always do the optimization of influencer matching and price matching. If OpenClaw comes to call, let it call; the dirty outreach work in the middle still has to be done by me. GTM hasn't changed either, because OpenClaw still has an owner, and the object of our Marketing is the owner of OpenClaw. Whether the decision-maker is a CMO or an intern, the person making decisions in the future will still be a human.
Zhu He: Mr. Wang, if in half a year Agent traffic exceeds human users, what do you think needs to be done to welcome it?
Wang Ming: Our system itself is Agent-to-Agent collaboration, using massive amounts of our own Skills. In the future, everyone may have a Personal "little lobster" (OpenClaw). All centralized entry points may disappear, meaning we want to become the Agent OS for a certain group. If OpenClaw solves the long-term memory problem, combined with the open-source ecosystem and its 5,000 PRs per month iteration, it will be the most important Personal AI open-source framework of the next era.
Zhu He: A friend in Silicon Valley said OpenClaw is the first thing to take back human memory from internet giants. Then Jiancong, if in the future users mainly use OpenClaw, will you degenerate into a database (Dataset)? Perhaps in the user's Skill combination, you are only responsible for being triggered by queries.
Wang Jiancong: There are several things in the whole chain that we don't pursue being the strongest at (like databases), but some things we must be the strongest at. For example, outreach, and the pricing model for influencers. The pricing model not only looks at static indicators like country and follower count, but also accurately judges follower profiles and geographic location. Additionally, it combines dynamic market conditions—like "when a typhoon comes, does ride-hailing raise prices?"—this kind of floor-price model combined with historical collaboration data is something no pure Skill has.
Zhu He: If a client only needs your bidding capability and splits you into several pieces for API calls, do you support that?
Wang Jiancong: I don't support it. This is like ride-hailing: my service is to bring the car to you. You can't say you only use my Agent to check a price but not take the car. As long as my Skill is differentiated, I have a chance to be used—we are not afraid of being different from others.
Zhu He: Anbang, you are the most Cloud Native—what are your next development plans? Want to become the next Hugging Face?
Xu Anbang: Hugging Face is mainly for humans, but many capabilities inside can be used by Agents. In the Agent era, the most important thing is how you use large models to do things—the "cognition." As for making tools, making data or making communities, those are all things at the "technique" level.
Zhu He: You have an open mindset on one side, and on the other side you use old methods to first earn that US$100,000—pretty impressive. Finally, everyone summarize in one sentence how to welcome future plans.
Wang Jiancong: You should embrace all changes. The cost of development and testing is really too low now—if you have an idea, just test it directly.
Xu Anbang: We are actually making a new-era Video Agent. Whether it is for humans or for Agents, we hope to become the number-one video creation entry point, empowering global creators and Agents.
Wang Ming: First, embrace the new traffic channels of every era; second, pay attention to new open-source technical frameworks, like OpenClaw. Finally, what never changes is returning to the essence of business: in the early stage, quickly cut in, get capital and users, build your own ecosystem position and capital moat, and become an entry point of some form in the next era.
Zhu He: That's about it for today. Everyone may think I am overly optimistic about Agent traffic exceeding humans. But the internet has hundreds of millions of users, mobile internet has billions, and in the Agent era it is at least on the order of tens or even hundreds of billions. Should everyone here start thinking about how our products can let OpenClaw use them for us? The future belongs to those who move first—thank you all!