DeepWisdom Raises $220 Million in One Year, Finally Launches First Product

"Not just writing code. Starting a company." By Muxin Xu

Edited by Zhiyan Chen

Another coding product hits the market.

Even amid the frenzy of agent investments, DeepWisdom's fundraising record stands out. In 2025, the company completed two rounds totaling 220 million RMB, with backers including Ant Group, Cathay Capital, Jinqiu Fund, Baidu Ventures, and Notion Capital.

DeepWisdom founder Chenglin Wu graduated from Xiamen University's computer science department and previously worked at Huawei and Tencent. His research has been published at top-tier conferences including ICLR, KDD, CVPR, and ACL. Core team members hail from Google, Anthropic, ByteDance, Tencent, and Huawei.

Now, DeepWisdom has officially launched its flagship product: Atoms (formerly MGX), a multi-agent AI coding platform built around the premise that "one person can have an entire startup team." Users summon AI agents — product managers, engineers, architects, researchers, data analysts — to complete the full product cycle from research and design to development and deployment.

Atoms' predecessor, MGX, shot to #1 on Product Hunt's weekly ranking after its February 2025 launch and broke $1 million ARR.

On the eve of Atoms' official release, Anyong Waves sat down with Chenglin Wu. "Turning every person's idea into a profitable product" is the core of his vision. This conversation covers how he defines AI coding, his exploration of multi-agent systems, the business model of coding products — and a radically transformed blueprint for society.

The interview follows —

Part 01

Not Writing Code, Starting a Company

Anyong: In the plainest terms possible, what is Atoms?

Chenglin Wu: Simply put, Atoms is a platform that lets you "launch a startup with one sentence." You have an idea, feed it to Atoms, and it summons a team of agents — product managers, architects, engineers, QA, data analysts — to research, design, develop, and deploy a complete product.

Previously, coding had a high barrier to entry. Non-technical people had almost no path from zero to a commercial tool. Now, as long as you have an idea, Atoms handles everything end-to-end: requirements research, feature definition, code development, user registration, payment integration, backend deployment — the whole package.

For example, one user wanted an online voting site. They just told Atoms, "I want a camping food voting page," and it automatically completed the research, page design, database configuration, authentication, and one-click deployment. You don't need to know code or backend systems. So we don't see it as a "tool" — it's a "cofounder."

Anyong: Compared to competitors like Lovable or Replit, what's your core differentiator?

Chenglin Wu: Lovable and Replit are primarily "code-writing" tools. Atoms is a "company-launching" platform.

Our multi-agent architecture directly enables full-stack capabilities, forming complete front-end and back-end development — user login, payment systems, databases, content management, API integration, the entire backend suite. On top of that, we're highly cost-efficient, using open-source models (like DeepSeek, Qwen) in combination at roughly 10% of GPT-4's cost.

Anyong: Google's Gemini 3 stunned everyone. Is that a threat to Atoms?

Chenglin Wu: Not at all — it's a tailwind. After integrating Gemini 3, Atoms' output quality improved significantly, and it brought us more users.**

And Atoms' positioning is fundamentally different from single-point models or coding products. It's not a coding assistant — it helps users rapidly launch a complete, commercially closed-loop business. Real-world complex commercial tasks require team collaboration; they're never accomplished by one large model or one individual alone. So Atoms functionally packages the entire business chain: market research, full-stack development (including user authentication, payments, database backend), deployment and operations, product marketing, data analytics, and more. While other AI products are still stuck at writing code and producing demos, Atoms has evolved from Vibe Coding to Vibe Business — directly delivering a complete, money-making operation. Therefore, model capability improvements just enhance our single-point capabilities like coding. It's like a company hiring a stronger programmer — within Atoms, that programmer collaborates better with other business functions, generating greater commercial value. That's even more favorable for us.

Anyong: Since MGX launched, has any user case stuck with you?

Chenglin Wu: Our user base is actually very diverse — e-commerce sellers, freelancers, outsourced developers, content creators, educators, vertical founders. So we see everything from small SaaS tools: one user built an AI image tool, plugged in Atoms' registration and payment features, and turned it into a paid business. To e-commerce sites: a Malaysian jeweler with zero coding knowledge built a polished storefront using Atoms, integrated Stripe payments.

What moved me most was an eighty-something Canadian grandfather with no technical background. He used Atoms to create a personalized education product for his granddaughter. He enthusiastically agreed to our interview and expressed deep gratitude that our product helped create so many interactive moments with her. This case reinforced my belief that Atoms doesn't just help completely non-technical people rapidly realize a product — it carries warmth and intergenerational connection, bringing people closer together.

Anyong: What's your current pricing and long-term business model?

Chenglin Wu: Currently subscription-based, billed by token usage and add-on features**; long-term, since we want everyone to turn ideas into profitable businesses and achieve minimum viable atomic entrepreneurship, the business model will revolve around commercial value delivery — platform fees, revenue sharing from user-integrated payments and advertising, infrastructure fees across the broader ecosystem network. In short, there's much broader commercial space ahead.

Part 02

The Endgame Is Multi-Agent

Anyong: You bet on AI coding very early. How did that conviction form?

Chenglin Wu: By late 2022, we saw natural language-to-code capabilities breaking through. We judged that AI coding would be among the first directions to achieve viable business models. Because the output is verifiable, the value loop is closed, and commercial value can be generated quickly.

Anyong: But the barrier to entry for AI coding was still quite high back then.

Chenglin Wu: Yes. So we built MetaGPT from the start — the world's first multi-agent coding framework. We wrote a 400-page paper to define the foundational agent technical roadmap. Now our Foundation Agents has surpassed 150,000 stars on GitHub, making it the world's largest multi-agent open-source organization.

Anyong: You keep emphasizing "multi-agent." Why so convicted?

Chenglin Wu: The future business world will be composed of countless "AI-native companies." Because real-world complex tasks are never completed by one person — they're accomplished through team collaboration. Think about it: how many roles coordinate from research, product design, architecture, development, testing, deployment, to operations for a single piece of software? Large models process information; agents process "goal-driven behavior." We believe the future is an "agent internet" — everyone has their own agent, not just a chatbot but an AI avatar that executes tasks, collaborates, communicates, and achieves objectives.**

We're even pushing for an "agent protocol" — just as the internet had HTTP, agents now need a communication protocol between them. We're inviting the world's top professors and institutions to work on this together.

Anyong: Your company seems to have a very academic atmosphere.

Chenglin Wu: I want to build an "academic loop" organization. Internal documents, plans, outputs are nearly completely transparent. Every task must have atomic deliverables. Performance is determined by peer scoring. OpenAI, DeepSeek, early ByteDance all followed the "academic loop" principle. So do we.**

Anyong: Does this approach actually work for entrepreneurship in the commercial world?

Chenglin Wu: Because I believe innovation isn't accidental — it's determined by organizational structure.**

We're building an "academic loop" organization to pursue extreme innovation capability, which is the only business moat in the AI era. It has several benefits. First, accelerated innovation: with nearly transparent internal documents and plans, and atomic deliverables for every task, information flows efficiently without departmental silos, enabling rapid iteration of technology and product roadmaps. Second, extreme efficiency: we established a "root group" — a team of geeks collaborating with AI agents, end-to-end identifying and solving problems, bypassing lengthy requirement handoffs. Have an idea? Just execute directly. Traditional companies are slow and conservative. We believe this organizational structure lets us fully embrace AI. So this approach doesn't just work — it's our fundamental guarantee for sustained leadership in the AI era.

Anyong: What's your current team size?

Chenglin Wu: We're 80-plus people, split between Xiamen and Shenzhen, planning to expand to around 120. We look for two things in hiring: technical ability and mindset. We want everyone to have critical thinking and make atomic contributions.**

Anyong: With more "super individuals" and "one-person companies" emerging, will the multi-agent path accelerate this trend?

Chenglin Wu: The future business world won't be monopolized by large corporations. It'll be an ecosystem network of countless "atomic" small companies. Everyone can use platforms like Atoms to turn their ideas into products, services, sites, content, and monetize them. No need to register a company, hire people, or raise funding — just have an idea, build an MVP, iterate gradually. We call this "atomic entrepreneurship."

This is also the meaning behind our company name DeepWisdom: unlocking the 97% of human creativity and data that currently lies dormant in society, releasing it through the agent network as new commercial value.

Anyong: What happens to traditional companies then?

Chenglin Wu: They'll become very slow and conservative. Large companies resist change; mid-sized companies dare not use AI to manage people. Individual entrepreneurs, by contrast, have short decision chains, embrace AI more thoroughly, and seize opportunities more easily.

Anyong: If this future you describe actually comes to pass, what role do you hope to play in it?

Chenglin Wu: I don't want to make tools. I want to build the infrastructure for the entire "agent internet."

Part 03

This Generation of Entrepreneurs

Is Supplying "Intelligence" Itself

Anyong: Your 2025 funding scale ranks among the top in this sector. How did you achieve that?

Chenglin Wu: Last year we completed two rounds: in March, Ant Group invested 100 million RMB; in June, a USD round led by Cathay Capital with Jinqiu Capital and others participating, totaling $17 million. Combined, that's 220 million RMB. We were 4x oversubscribed. We actively chose not to take more, to avoid inflating valuation too early and constraining future growth room.**

Anyong: Is fundraising ability the biggest moat for AI startups now?

Chenglin Wu: Investors oversubscribed primarily because they bought into our long-term vision and technical roadmap. Fundraising ability itself isn't the moat — technical depth, product efficacy, endgame conviction, and organizational efficiency are.

Anyong: What do you think investors bought into?

Chenglin Wu: Our long-term vision and technical roadmap. Many say we're the best in open source, and we saw "multi-agent + AI coding" as the future trend earliest. Our Foundation Agents has surpassed 150,000 stars on GitHub, making it the world's largest multi-agent open-source organization. Investors recognized us as "the best in open source." Beyond that, our ability to deliver complete commercial value and extreme engineering capability were also key reasons. Efficacy and cost are critical metrics in this赛道. Our own benchmark: European and American competitors average 0.4-plus; we've achieved 0.8 to 0.9-plus. Atoms delivers 80% lower cost than competitors while exceeding their efficacy by 45%. This efficient, low-cost delivery capability brings higher certainty and return potential for capital.

Anyong: Things are going well now, but I heard you nearly went under three times?

**Chenglin Wu: Yes. Between 2021 and 2024, we had three cash flow crises. We survived on client payments, friends' loans, and team salary cuts. We never missed payroll. Core members went three years without raises or year-end bonuses, and most stayed.

Anyong: Any regrets?

**Chenglin Wu: Often (laughs). I had an offer in 2017 — if I'd taken it, I'd have 200 million in options by now.

**But actually, I knew I wanted to start a company since college. Early on I tried AI stock trading, B2B services. But what solidified my direction was realizing that everyone's ideas have value, but there was no way to realize them. We want every person's idea to be taken seriously, rapidly realized, and rewarded. That's the original intention behind Atoms, and our most important mission for the coming years.

Anyong: Is there a fundamental difference between this generation of AI entrepreneurs and the mobile internet era?

**Chenglin Wu: The last generation of CEOs was resource-driven; this one is technology-driven. We're not accelerating information transmission — we're supplying "intelligence" itself. This requires completely different knowledge systems and organizational forms.

Anyong: Elaborate?

**Chenglin Wu: For the past 70 years, including the mobile internet era, productivity expansion in the business world fundamentally relied on human brain scaling. Organizations expanded through hierarchy, products through manpower, startups through headcount growth. Core capabilities were "stacking people" and "stacking resources." But today, silicon speed is improving exponentially, costs dropping exponentially. For the first time, we have massively scalable "silicon intelligence." This means future productivity expansion won't depend on hiring, but on silicon scaling. With sufficient compute, your "team" can expand infinitely. Silicon speed improves exponentially, silicon costs drop exponentially, meaning: future productivity no longer depends on human brain scaling, but silicon scaling. Once intelligence is no longer scarce, the boundaries of teams, the definition of companies, the cost of entrepreneurship — all get rewritten.

**Many current AI products are still stuck in the old "human brain scaling" mindset, just helping people draft copy or produce demos, without truly leveraging silicon scaling to replace commercial links themselves. What we mean by supplying "intelligence itself" is bridging this gap: making silicon not just generate content, but complete the entire commercial chain from research, decision-making, development to revenue for you. We named it Atoms with this meaning — every fragmented idea is the minimum unit before scaling. Through the Atoms product, we expand the boundaries of teams, companies, and entrepreneurship, achieving a production revolution in the new commercial society.

Anyong: As CEO of a multi-agent coding company, what percentage of your daily work involves agents?

**Chenglin Wu: Not much yet, but it's already saved me enormous time on research, SQL writing, and report drafting. In the future, I hope it becomes my "thinking extender," maybe even give interviews for me, haha.

---

Recommended Reading

Where Money Flows, People Rise and Fall