Original · Unique Research · 2026-03-22 · Shanghai
Editor's note: This is the complete English rendition of Unique Research's March 22, 2026 interview-based article. The first-person commentary is the original author's. Product Hunt placement, the description of OpenClaw as the world's largest agent ecosystem, its 300,000 GitHub Stars, the 90% figures, product-performance comparisons, integrations, preference learning, and commercialization claims are statements by the source or interviewee, not independently audited findings or current measurements. Predictions for the end of 2026 and the next 12 months remain historical predictions. The introduction calls Steven a founder and the interview credit calls Steven Cen a co-founder; both source formulations are preserved. Descriptions of learning team preferences do not establish privacy, consent, or data-governance safeguards.
Unique Awards · Guest Interview
Giving Up Technical Lock-In, This AI Team Found an Even Stronger Moat
In the AI era, functional moats have already collapsed.
If Cursor can copy 90% of your core features in a weekend, does your AI product still have a moat? ChartGen founder Steven has a clear-eyed answer: "In the AI era, functional moats collapsed long ago. What you can actually defend is preference lock-in." This article explores where ordinary entrepreneurs can find barriers to competition as the AGI wave penetrates the application layer. Anyone building a product should read it.
If an independent developer can copy 90% of your product's core features with Cursor over a weekend, how can you keep running that business?
Worse still, major players such as Microsoft's PowerBI Copilot and Salesforce's Tableau AI are watching from right beside you.
It is no exaggeration to call this the suffocating dead end facing every AI application-layer entrepreneur this spring.
But in a recent conversation, ChartGen AI co-founder Steven Cen offered a strikingly counterintuitive take:
"In the AI era, functional moats have already collapsed. The barrier that is genuinely hard to copy is preference lock-in."
To be honest, that phrase stopped me for a moment. This was a rising team that had taken first place on Product Hunt's daily ranking just last year and was providing native visualization capabilities for OpenClaw, the world's largest agent ecosystem. Surely some proprietary algorithm was what they were proudest of?
With that question in mind, I took the conversation further.
What emerged was that this local battle over "who turns data into charts" had already become a compass for anticipating the next generation of software. It explains not only why so many AI tools end up as cannon fodder, but also what asset can never be stolen in a future where agents take over entire workflows.
When AI Cuts Through Everything, Who Is Still Serving the Tools?
Let me start with a stupid question: what does traditional data analysis feel like?
You buy a prohibitively expensive BI system, take training classes to learn how to connect databases, and then memorize a pile of obscure drag-and-drop rules and formulas. To understand a single row of data, people have to serve the tool unconditionally.
"90% of people in enterprises cannot write SQL or Python," Steven told me as he laid out the situation. Yet those very same 90% have to struggle through reports on business growth and channel ROI every day.
The pain point is so obvious that countless teams are building conversational AI that "generates a chart from a sentence."
But ChartGen has found a more incisive route: let go of the obsession with "building for people" and become the "eyes and paintbrush" of AI agents.
"Data itself has no value. Data that can be understood and communicated does."
Over the past month, everyone has witnessed the explosive growth of the OpenClaw operating-system ecosystem: 300,000 GitHub Stars and thousands upon thousands of community capability modules. Enterprises are getting used to having large models collect and clean data.
But at the final step, when a report has to be produced, something awkward happens.
"The traditional data-visualization module built into OpenClaw can only produce those incredibly ugly charts in Matplotlib's default style," Steven said pointedly. "They work, but they don't look good—let alone good enough to present to your boss."
Imagine that your future customers are no longer human office workers, but rapidly running agent workflows. How would you earn money from them?
ChartGen's answer is to embed itself in those workflows. From the first day of architectural design, the team did not view the product as traditional SaaS. Instead, it packaged it as API skills that OpenClaw and a variety of intelligent systems in China could call directly.
While others compete for shortcuts on human desktops, they are competing for a place as infrastructure in agents' chains of thought.
This change in perspective has taken ChartGen out of the crowded battle among traditional BI tools over reporting features.
80-Point Automation and 100-Point "Preference Lock-In"
If you want to become an agent's paintbrush, surely the technology can still be copied? Surely a major company can build its own module?
At this point in our conversation, Steven finally revealed his real trump card:
"General-purpose large models will cut straight through features built on if-else logic. But features that require human aesthetics, collaboration, and intervention are very difficult to cut through directly."
That was the line that struck me most in the entire conversation.
What limits a fully automated agent? Ask AI to draw a chart of a marketing budget, for example, and it will quickly produce a perfectly serviceable pie chart. That is an 80-point product that meets basic needs.
But in the real business world, that is nowhere near enough.
Your director may look at the chart and say: "Highlight that channel with the terrible conversion rate in bright red. We're going to call it out at this week's meeting." The chart then goes to a designer, who says: "This color scheme is tacky. Use the company's brand VI palette." Finally, the boss looks at it: "Add an annotation to the bar at the top left. We reorganized that business last month."
No fully automated agent can simply guess its way through this kind of collaboration, full of complex interpersonal intentions and subjective aesthetics. The major companies' mechanical one-click generation can, at best, reach what Steven calls "Competent but mediocre"—usable, but not impressive.
So ChartGen has made "Human-in-the-Loop" the central bet of its product.
Once a chart has been generated, team members point, mark it up, and edit it together. In that process, the engine is learning furiously: oh, this team's boss cares enormously about conversion rates; oh, their brand guidelines never allow bright yellow.
The so-called "collaboration flywheel" generates a substantial store of preference data.
Compared with traditional software's "technical lock-in," which forcibly ties users down through licensing agreements, the more insurmountable moat is a system that understands you better the more you use it—and the feeling that leaving would mean losing even your unspoken mutual understanding.
That is preference lock-in.
The Open-Source Hook and the Closed-Source Assassin
The logic sounds complete. But in today's model frenzy, which prizes free access and open source, how do you make real money?
That brings us to another harsh truth about the agent ecosystem: if your component is not open enough, developers will not even look at it.
"ClawHub is like npm for agents. If your skill is awkward to use, people will uninstall it and switch to another in no time."
To claim the default first-choice position in this jungle, ChartGen has chosen a very smart asymmetrical approach: an open-source ecosystem and a closed-source core.
It has generously open-sourced data-visualization interaction standards and interface code for the OpenClaw community, along with best-practice documentation.
The hooks cast out are all useful, free, open-source material, earning the team reputation and traffic. This also leaves closed-ecosystem competitors from Microsoft and Salesforce without an entry point into this vast, untamed community.
But firmly attached to the other end of the hook is ChartGen's uncompromising closed-source blade: a proprietary design-intelligence engine, professional color algorithms, and the preference-learning system that keeps people coming back.
Consumer users get a free trial. But once enterprise agents settle into habitual use and call visualization tasks hundreds or thousands of times every day, they cross the threshold into paid subscriptions quite naturally.
That is how you use an open-source exterior to make money from a complete commercial cycle.
Epilogue: Making Room for Human Intelligence
At the end of our conversation, I asked for his prediction of the market at the end of 2026.
The future will most likely split into two entirely different quadrants.
One path leads into an intensely competitive race for maximum automation: who computes faster, and who charges nothing. Ultimately, these products become indistinguishable basic utilities.
The other path looks more like ChartGen's: knowing when to step back half a pace at the most fiercely contested point. Hand the 80 points of laborious work to agents for fully automated processing at scale, and return the 20 points of bespoke experience that determine survival to humans who provide preference feedback.
The ultimate destination of tools has never been the complete replacement of human hands. Only living people care whether the presentation behind a row of numbers is beautiful or compelling.
The next time you want to add another "killer automation feature" to your AI product, pause and ask yourself:
What are you desperately trying to defend: a feature, or people's hearts?
Selected Q&A
Q1: What kind of company is ChartGen today?
ChartGen currently has two main lines of business. One is a SaaS platform for overseas end users, who can upload data or connect databases and use natural language to generate charts, dashboards, and PPT presentations. The other is a skills module for the agent ecosystem, providing professional data-visualization capabilities for frameworks such as OpenClaw and ClawHub.
Q2: Why did Steven choose AI charts and data visualization?
The initial trigger came from work on Ada.im. The team found that after asking AI a data question, users would always end by requesting "turn this into a chart" or "export this as a PPT." That showed that what users really needed was not a piece of analytical prose, but a result that could be understood, conveyed, and used to communicate.
Q3: What is the biggest difference between ChartGen and traditional BI tools such as Tableau and PowerBI?
Steven sums it up in three phrases: instant, no barrier to entry, and agent-native. Traditional BI tools require people to learn, configure, and drag and drop. ChartGen is closer to "describe it and generate it." It has also considered agent-call scenarios from the outset: not just an interface for people to use, but a capability module for agents to call.
Q4: Why say general-purpose agents cannot produce high-quality data visualization?
It is not that they cannot do it, but that most can only reach the level of "usable." General-purpose agents can produce charts, but often fail to make them presentation-ready: ready to use directly in reports, client meetings, or senior-level decision-making discussions. Truly high-quality charts must balance professional communication, colors, annotations, brand guidelines, and information structure.
Q5: What capability does Steven consider hardest to copy?
Not chart generation itself, but design intelligence, feedback data from the agent ecosystem, brand recognition, and the preference-learning system developed through Human-in-the-Loop collaboration. The last of these, in particular, enables the system to understand a team's style of communication increasingly well. Copying a few sections of code cannot achieve that.
Q6: How do you balance open source and commercialization?
Steven's strategy is "open-source ecosystem, closed-source core." Skills modules and best practices are open-sourced wherever possible to make it easy to enter the developer ecosystem. But the design-intelligence engine, AI understanding layer, color algorithms, and preference-learning system that create real differentiation remain on the closed-source server side.
Q7: Which opportunities does Steven find most promising over the next 12 months?
He is most optimistic about the Agentic Operating System ecosystem itself, especially financial agents, operations agents, and IoT agents. These scenarios share a common feature: once an agent genuinely enters enterprise workflows, it will inevitably need professional data-visualization capabilities.
Q8: What is his view of the market at the end of 2026?
By the end of 2026, data visualization will divide in two directions: bulk visualization executed automatically by agents, pursuing efficiency and scale; and premium visualization created through Human-in-the-Loop collaboration, pursuing aesthetics and expressive power. Products with real opportunities will serve both worlds.
This article was compiled from an in-depth interview with ChartGen AI co-founder Steven Cen.