Yunqi Capital Backs CREAO AI's Pre-A Round with Tens of Millions of Dollars to Rebuild the Work Entry Point with Agent OS | Yunqi Portfolio

CREAO AI, an agentic AI startup and Yunqi Capital seed-round portfolio company, recently announced a new funding round in the tens of millions of USD, with Yunqi Capital doubling down in this round. Founded less than a year ago, the company has raised over $30 million in total funding and surpassed 200,000 users — almost entirely through organic growth.

While most AI startups are still competing on model capabilities, CREAO AI has chosen a different path: rather than building a "smarter answerer," it is constructing a self-sustaining agentic loop — from generating tools, to invoking tools, to executing tasks autonomously. This edition of "Yunqi Partners" takes you inside the details.

The following content is from "Z Potentials"

Most AI startups sell a better model. CREAO AI sells a better loop.

The company has just completed a funding round in the tens of millions of USD, led by Prosperity7 Ventures — the diversified venture fund under Aramco Ventures — and Matrix Partners China, with participation from existing investors including Yunqi Capital, Monolith, Hillhouse, HSG, and Hua Capital. In less than a year, CREAO AI has cumulatively raised over $30 million.

The core thesis they are betting on: the real bottleneck in AI is not intelligence itself, but the gap between "a chatbot answers a question" and "an agent works for you while you sleep."

Two Traps

"Everyone knows AI will unleash productivity," says CEO Kai Cheng. Before founding CREAO AI, he spent ten years building production-grade AI systems for over 250 enterprise clients. "But the entire industry is stuck in two traps: if humans still have to operate AI tools step by step, productivity has a ceiling; if humans remain the only ones building tools, the real AI revolution hasn't even begun."

CREAO AI's solution is a closed-loop system: AI both builds tools and runs them.

This loop materializes as a complete product architecture:

CREAO starts with a conversation. Users describe a task to a "super agent" — it doesn't just answer questions, it executes directly: writing code, calling APIs, connecting third-party services, and delivering results in a sandbox environment. The key is what happens next. A successfully executed task is saved as a reusable Agent App — a persistent, auto-executing unit with independent memory that can trigger on schedule. That SEO pipeline you ran on Tuesday? It runs again on Friday. Without you being there.

The product architecture has three layers:

Coding Agent

AI builds tools. Users create Agent Apps through conversation, without writing traditional code. The team calls this breaking the "builder bottleneck."

Autonomous Execution

AI uses tools. Agent Apps run on schedule, automatically triggering workflows without human follow-up. This breaks the "operator bottleneck."

Workspace

Humans command. A persistent environment where Agent Apps run and memory accumulates across tasks. One person can manage workloads that previously required an entire team.

Stacked together, these three layers form CREAO's Agent OS — an agent-centric operating system entry point covering all work scenarios. In the future, whether it's content production, customer operations, data analysis, or code delivery, everything will launch from here, accumulate here, and be driven by here.

"If AI builds tools but humans still have to click 'run' every time, we haven't won," says Cheng. "The core is the closed loop: AI builds, AI runs, humans steer."

Five Pivots to Find Direction

This direction wasn't visible to the team from day one. It was found through trial and error.

When CREAO AI launched in September 2025, it was a Vibe Coding tool. By December, the entire product was torn down and rebuilt around the Agent App model. But this pivot was only the most recent in a series. Since assembling in late 2024, the team has experimented with synthetic data, workflow builders, and natural language programming, until the "super agent" approach finally clicked.

"We kept thinking the problem was something specific — data, workflows, code — but each time we discovered the real issue was deeper: how do humans and agents actually collaborate," says co-founder Clark Gao, who leads GTM (go-to-market) and previously built data teams at LinkedIn and Tencent.

What makes this pivot story unusual is the density of the team's credentials. CTO Peter Pang was a research scientist on Meta's Llama 3 team — one of the most influential projects in open-source AI history — and before that, did multimodal model research at Apple. When he joined, he brought a judgment that others found somewhat exaggerated at the time: "This is just the starting point of a new evolution. The speed will exceed anyone's expectations."

Peter established an AI-First engineering culture at CREAO. In a recent internal memo, he wrote: "AI won't diminish the value of engineers; it changes where that value lies. Engineers' value will no longer be measured by lines of code written, but by clarity of thought and quality of decision-making."

Execution Is the Moat

Cheng has an unusual view on strategy: he believes strategy matters less than people think right now.

"AI is evolving too fast. What looks promising today could be identified and copied by competitors tomorrow," he says. "Product direction at the conceptual level is no longer a defensible advantage."

He uses a manufacturing analogy. "Why is it hard for competitors to surpass China in hardware manufacturing? It's not conceptual leadership in design, but massive structural efficiency advantages in the supply chain. For AI startups, there is only one real strategic advantage — who can adopt AI first, internalize it, and use it to amplify their own efficiency by 100x."

CREAO is its own heaviest user. The company's KOL marketing, SEO pipelines, and content production all run on the CREAO platform. A team of fewer than 20 people outputs workloads that would traditionally require several times the headcount.

One agent replaced a three-person SEO workflow overnight — keyword research, copywriting, page design, deployment, fully automated end-to-end. Another agent ran a content pipeline for two days before someone noticed the output was garbage. Both happened in the same week. The team calls themselves "crash test dummies for the future of work." It's not entirely a joke.

"Every internal failure becomes product insight," Cheng says. "Every broken workflow becomes a feature requirement. We have to first become the ultimate AI-driven company internally before we have the right to sell this to the world."

A Hot Sector, A Different Bet

The AI agent market is projected to reach $52 billion by 2030, with a CAGR exceeding 46%. Capital is pouring in.

Manus was acquired by Meta for $2 billion; Genspark raised $275 million in Series B at a $1.25 billion valuation; Gumloop raised $50 million, focusing on no-code enterprise automation; Relevance AI raised $24 million, offering pre-configured agent teams.

CREAO AI is riding the same wave, but its bet is fundamentally different. Gumloop lets users drag and connect visual workflows on a canvas; Relevance offers pre-built agent teams; CREAO skips the construction interface entirely — no canvas, no drag-and-drop, no node editor. You speak, the agent executes, and the result directly becomes infrastructure.

"Everyone else is building better ways to command AI," Cheng says. "We're building the layer where AI work compounds — every run feeds memory to the next, and successful runs become self-running apps."

Manus proved that a single powerful agent can complete complex tasks. CREAO is betting on the next step — making that capability persistent and repeatable. Genspark built a workspace for knowledge workers; CREAO's workspace is designed for orchestration — not for humans to use AI tools, but for humans to manage autonomous agents that use tools on your behalf.