Original · Unique Research · 2026-07-11
Editor's note: This is the original Chinese author's interview with Li Jinwei and its framing. This English rendition retains the full text in source order, including all 12 Q&A items. The interviewee's product claims, growth figures, and market views are his self-reports, attributed to him and not independently verified. Person, company, and product names are preserved as source attributions.
AI Industry Observation
Leave the "doing hands" to agents; leave judgment and taste to humans.
In Spain, a few dentists recently got hooked on a tool. They don't know design, can't use Photoshop, and have no aesthetic training. But with one sentence, they generated a patient-facing clinic notice, printed it and taped it to the clinic door — clean, professional, no worse than any brand clinic's materials. The same scene happened inside an enterprise: someone fed a whole long block of corporate-culture and policy text in, and minutes later got a visual long-image ready to mass-forward.
This tool is Open Design. We interviewed its CMO, Li Jinwei (李锦威).
95% of the Time Is Wasted on Things That Don't Create
First, a question: how long does a shippable product page take? The answer is usually — two weeks. But Li Jinwei gave a starker breakdown: in those two weeks, the truly creative part — what this thing should actually look like — is under ten minutes. The other 95% of time is spent finding designers, aligning in meetings, revising, slicing assets, back-and-forth communication. This isn't "optimization room" for an efficiency problem; the whole process is built wrong.
What Open Design wants isn't to compress those two weeks into one and a half, but to directly replace the work unit: from "delivered in two-week units" to "delivered in units of one intent, a few minutes." You say "make me a Stripe-style landing page," and minutes later you don't get a reference image but dozens of components, a whole design system with tokens, one-click export. This isn't a quantitative change; it's a different species.
Most People Understand It Wrong From the Start
People seeing Open Design for the first time almost instinctively classify it as "another AI design tool" — something like Midjourney or Canva AI. Li Jinwei says this is the biggest misunderstanding. Most AI design tools on the market are essentially "generating images": a prompt in, an image out, and that image is what it understands as "design." But the image can't be used directly — you still turn the image into HTML, then HTML into shippable code; every conversion loses and distorts.
"Others are drawing pictures; we're directly manufacturing."
Open Design never intended to draw pictures from the bottom up; the moment it generates, what comes out is already deliverable production code — both design mockup and final product. That's why, in their definition, "generating an image" never counts as done. The true completion bar is: usable, editable, deliverable. It can go live directly, support versioning, and let frontend, product, and testing keep collaborating on the same code. The image is only the starting point; what can actually be used is the end.
Why Was "Design" Chosen
AI can do many things — write code, write copy, do customer service. Why did Open Design bet on design? Because a particularly typical gap opened here, with two curves colliding at once. First curve: coding agents are fully exploding. Cursor grew from an IDE plugin into an independent track valued at $60 billion; AI open-source projects on GitHub rose from 400,000 three years ago to an expected 1.2 million this year. The barrier to making products was smashed to the floor. The second curve, however, is that design and taste iterate far more slowly. More and more people can make things, but very few can make them look good and on-brand. And something that happened in 2026 made this gap fillable for the first time — models truly acquired taste for the first time. After the Opus generation, models can already write world-class beautiful interfaces by default.
On one side the most painful gap, on the other a just-mature technology inflection point; when the two meet, the answer is almost inevitable.
It's Not a Tool but a Partner That "Understands You More the More You Use It"
Strictly compared with Midjourney, Canva, Figma AI, these aren't the same thing — they're essentially still the old logic of "a person manually operating inside a tool," just differing in how smooth the operation feels. Open Design is more like a design partner that does the work itself and understands you more the more you use it. Its focus was never to make you "operate more smoothly," but simply to do the work for you.
This "understands you more the more you use it" takes two concrete steps. First, quickly build your brand-asset library. You don't need to teach it from scratch what your brand looks like; just throw in your website link, open-source repo, Figma file, and it reverse-extracts a whole set of brand colors, fonts, spacing, and component patterns into a systematic design system. Second, continuously evolve in use. If you iterate a page for thirty rounds, those thirty rounds of tuning settle into your team's own design system — next time you generate, the flaws you already fixed won't appear by default.
"Once taste can be remembered, it can be 'transferred' to cheaper models. If good design no longer depends on expensive models but on accumulable, reusable 'memory,' then 'looking good' can truly be scaled and democratized."
Who Uses It? Two Ends Stuck by the Same Thing
Open Design's users fall into two almost non-overlapping groups. One end: ordinary people and indie developers with no design background. The Spanish dentist mentioned earlier is one example; another overseas user originally knew nothing about design, used it to build a complete website for their product from scratch and ship it, and afterward voluntarily became an official overseas promoter. Most have no local agent; they knew the product through creators on YouTube and Instagram and casually use it as a daily visual-expression tool. The other end: big-company design and engineering teams — already in use at a top phone maker, a top content platform, a global tech company, and an AI coding company, with the list growing. These two ends look unrelated, but the stuck link is the same: design execution. One end can't do it; the other is dragged too slow by process. And for all users, what they truly get was never a "mockup" but something directly usable — a shippable page, a runnable prototype, a deliverable deck. The fastest-growing scenarios now are landing pages, product prototypes, social posters, decks — the kinds that "must look good and go live directly."
Zero Ad Spend, They Reached Nearly Ten Billion Impressions
Open Design now has over 70K stars. In GitHub history, only about three hundred projects have ever passed 60K stars; reaching this scale in such a short time is roughly top ten globally. But Li Jinwei doesn't treat stars as the most important metric. What truly supports this number is nearly 400 contributors from 30-plus countries, and nearly ten billion in organic impressions across the web — all without spending a cent on ads. The logic is plain: product good enough → what users make is both good-looking and usable → they naturally post work to X and communities → that itself is the best distribution.
"GitHub has in some sense become the Xiaohongshu of the AI era."
If he had to pick one metric that best represents this open-source product's vitality, Li Jinwei chooses not stars but contributor count. Because that's the one thing "others can't copy." You can buy 400 contributors' short-term input with money, but not their long-term, spontaneous, sustained contribution. Once a product truly spreads to hundreds of countries with thousands spontaneously co-building, that itself is a moat.
Endgame: Design's "Last Mile"
Pull the view further back; Li Jinwei's judgment is: writing code has been largely automated by agents, and design is the next link to be pulled into this automated loop — from Coding AGI toward Design AGI. Tools that purely sell "canvas" will be gradually hollowed out. Because their core value is "let you manually operate more smoothly" — but when the "doing hands" themselves are taken over by agents, this value can't hold up. What truly survives will change from "tool" to "capability platform": offering no longer a canvas but a set of capabilities any agent can call, that self-evolve and are co-built by the whole ecosystem.
So, if one sentence sums up what Open Design truly wants to change, the answer isn't "design" itself, but: who is qualified to make beautiful things. In the past, world-class taste was locked in the hands of a tiny few top designers, a scarce privilege. Now, when models for the first time have taste, and can remember and pass it on, "looking good" can for the first time become a default everyone can call. Leave the "doing hands" to agents; leave judgment and taste to humans.
Selected Interview Q&A
Q1: How would you explain Open Design to someone who has never touched it?
Li Jinwei: Design is going through a paradigm shift, from Figma to Open Design. Over past decades, design logic hasn't changed much: you open a software, drag shapes out pixel by pixel on a canvas. What we want is for design to no longer be "a software you have to open specially" but "a capability any agent can call at any time," like writing code today, melting into the whole production flow. More plainly: you just say in one sentence what you want, and minutes later you get a good-looking final product that goes live directly. It covers a wide range: brand-compliant landing pages, product prototypes, social posters, launch decks, even an animated product video, an interactive online page — all generated directly. Through the whole process, you don't need to find a designer, nor drag rectangles on a canvas yourself.
Q2: Many understand Open Design as an AI design tool or asset library; where's the biggest偏差?
Li Jinwei: The biggest偏差 is that "generating images" and "manufacturing" are two things. Many AI design tools essentially use a prompt to generate an image, and that image is what it understands as "design." But the image can't be used directly; you turn it into HTML, then HTML into shippable production code, and every conversion loses and distorts. Open Design takes a completely different path. We were born designing with code itself, so the moment we generate, what comes out is already deliverable production code. It's both design mockup and final product. In one sentence: others are drawing pictures; we're directly manufacturing. Another common misunderstanding is treating Open Design as a "tool." It's actually more like a self-evolving agent; the more you use it, the more it understands you. Also to clarify, Open Design itself is completely free; as long as you have your own agent, like Claude Code, Codex, Cursor, Hermes, you get the full-link capability. What we charge for is only the additional official agent we offer, out of the box, for users who don't yet have their own agent.
Q3: Does Open Design truly solve "design faster," or "let people who can't design also produce professional results"?
Li Jinwei: We want to solve both, but more fundamentally it's a capability issue. Today, making a production-grade page that goes live directly still takes about two weeks. But the truly creative part of this chain — what this thing should actually look like — takes under ten minutes; the other 95% of time is spent on communication, scheduling, manual execution, and back-and-forth alignment. Eliminating this 95% is efficiency-level value. But the more essential layer is: in the past world-class taste was a scarce resource locked in a few top designers' hands, but now with a model plus a design system, we can turn "looking good" into a default capability anyone can call. Let people who couldn't draw a rectangle also make professional things. That's what we truly care about; we call it taste democratization.
Q4: In your definition, when is a design truly done?
Li Jinwei: It must be usable, editable, deliverable to count as truly done. Generating one image isn't done, because it can't go live, nor be further modified and collaborated on. Our completion bar is outputting deliverable production code: it goes live directly, supports versioning, and lets frontend, product, testing, and growth keep collaborating on the code. Generation is only a starting point. Truly delivering and being used completes the loop.
Q5: AI can do many things; why did you choose design and content production as the first breakout scenario?
Li Jinwei: Because design happens to split open an especially painful gap. Two things happen at once these two years: one is coding agents fully exploding, like Cursor growing from an IDE plugin into an independent track valued at $60 billion; the other is AI pulling the "make a product" barrier straight to the floor, with AI open-source products on GitHub rising tenfold in three years, from 400,000 to an expected 1.2 million this year. Product count and dev speed both explode, but design and taste iterate far more slowly. The result: more people can make things, but very few can make them look good and on-brand. Plus something essential happened in 2026: models truly acquired taste for the first time. After the Opus generation, it already writes world-class beautiful interfaces by default. One side the most painful gap, one side a just-mature technology inflection point; these two together pointed us to design.
Q6: Compared with Midjourney, Canva, Figma AI, what is Open Design more like?
Li Jinwei: Strictly, none of them. Midjourney generates images, Canva is templates, Figma is a canvas; they're essentially still the logic of "a person manually operating inside a tool." Open Design is more like a design partner that does the work itself and understands you more the more you use it. Its focus isn't making you operate more smoothly, but simply doing the work for you.
Q7: Who are Open Design's core users — designers, marketers, founders, or ordinary users who can't design at all?
Li Jinwei: Our users actually fall at two very different ends. One end is ordinary users and indie developers with no design background. For example, an overseas user originally knew nothing about design and used Open Design to independently build a complete website for a product from scratch and successfully ship it. Afterward he loved the product so much he voluntarily became our overseas promoter. There are also some unexpected scenarios. For example, some dentists in Spain use it to generate patient-facing clinic notices, print them and tape to the clinic door; also people in companies turning long blocks of corporate-culture or policy text directly into a visual image for internal sharing. The other end is big-company design and engineering teams. Already several top phone makers, top content platforms, global tech companies, and AI coding companies use it, and the list keeps growing. These two ends look very different, but the stuck link is the same: the "design execution" segment — either can't do it, or dragged too slow by process.
Q8: Why do you stress Open Design is Agent-Native, not just a smarter design tool?
Li Jinwei: Because this is a bottom-level paradigm difference, not adding an AI button to an old tool. In the past design was "a software you open," manually operating step by step on a canvas; Agent-Native means design becomes "a capability any agent can call at any time," directly melting into production flow. This starting point makes us different from the bottom up: what we generate is directly production code; we stay neutral to any agent runtime, don't lock to one vendor, and every layer supports your own key. A smarter tool still ultimately makes you do the work yourself; Agent-Native hands the "doing hands" themselves to the agent.
Q9: For an ordinary user, from "I want a design" to "truly getting a usable result," what's the most key change?
Li Jinwei: The most key change is the whole "work unit" being replaced. In the past delivering a thing was in two-week units: find a designer, drag rectangles in Figma, revise many versions, slice assets, then have frontend implement, often with rework in between, 95% of time on process and alignment. Now it's in units of one intent, a few minutes: you say "make me a Stripe-style landing page," and it produces dozens of components, a whole design system with tokens and Storybook, one-click export. From two weeks to minutes — this isn't optimizing the process by a few percent, but directly replacing the work unit itself; it's an order-of-magnitude change.
Q10: You often say Open Design "self-evolves"; what does this concretely mean?
Li Jinwei: It's two steps. First, quickly build the team's visual assets. You don't teach it from scratch what your brand looks like; just throw in existing things — website links, open-source repos, Figma files. It reverse-extracts brand colors, fonts, font sizes, spacing, and component patterns into a whole design system. Second, continuously evolve during use. For example, to deliver a page you iterate thirty rounds back and forth; we record all thirty rounds of tuning and settle them into your team's own design system. Next time you generate, the flaws you already fixed won't appear by default. There's a severely underrated point here. Because taste can be remembered, we can fully use a very cheap model, like DeepSeek's Flash, paired with this tuned taste memory, to make designs comparable to Claude Opus level. Flash costs nearly 100x or more less than Claude. If we can stably use such a cheap model to make top-tier web pages, and everyone can do it, then "good design" can truly be scaled. More interestingly, this memory can also transfer. Someone with especially good taste tunes up this "brain," which is all code-accessible data and can be poured directly to others for reuse.
Q11: From a growth view, what's Open Design's core flywheel?
Li Jinwei: Our core flywheel runs around open source and word of mouth; the logic is plain. We first make the product itself good enough; what users make with it is both good-looking and directly usable, so they naturally post work to X and communities, which itself is the best distribution. In today's era, GitHub has in some sense become the Xiaohongshu of the AI era. Because writing code gets simpler and simpler, people share their projects as naturally as posting a note; media and self-media also actively watch projects emerging daily on GitHub to introduce and film videos. This way, a good product plus real word of mouth brings more use and more contribution. As the ecosystem thickens, outputs get better, and it rolls round by round. I want to stress especially that this whole growth was built with almost zero spend. Our web-wide organic impressions are already nearly ten billion, without spending a cent on ads. To make this flywheel spin faster next, we'll focus on two things: first, for users who don't yet have their own agent, provide an out-of-the-box official agent; second, for users with team-collaboration needs, offer team and enterprise editions, so Open Design runs not just personally but truly at team and company level.
Q12: If in one sentence, what Open Design truly changes isn't "design" itself, but what?
Li Jinwei: What we truly change isn't "design" itself, but "who is qualified to make beautiful things." This used to be a few people's privilege, but now every person with ideas can do it. Fundamentally, we want to return time to taste itself. Let people stop spending life on dragging rectangles and revising many manual versions, but spend it where it truly matters — on every judgment and choice of beauty.