Former miHoYo Executive's Startup: AI-Native Growth Agent Secures Funding, Core Thesis Is "Set Goals, Let the Machine Handle the Rest"

Event Background

LeapMind Growth, founded in 2025, announced the completion of its angel round and launched its flagship product GrowthGPT—an AI agent claiming to "independently take over the entire growth lifecycle." Users only need to set targets and budget boundaries, and the system automatically completes the full cycle of data diagnosis, creative insights, and execution optimization. The company was founded by growth leaders from mega-platforms including former Mihoyo, ByteDance, and Kuaishou, with the core team holding 50+ billion dollars in growth project experience.

Why This Matters

Traditional growth work is human-driven: data teams analyze problems, content teams iterate creatively, operations teams execute campaigns, backend teams optimize continuously—each step requires human judgment and decision-making. Even at mega-platforms, the bottleneck in this cycle is rarely a lack of data or tools, but rather the speed of human decision-making and error tolerance.

GrowthGPT's core innovation isn't a better algorithm—it's a redistribution of authority structure—pushing down the middle-layer decision-making that humans used to handle to machines, leaving humans with only two high-level decisions: "setting targets" and "defining boundaries."

From China Merchants Bank's automated credit approval, to Amazon's ad placement algorithms, to OpenAI's o1 model's increasingly autonomous reasoning, we see the same pattern repeating: the benefits of autonomy (speed, consistency, scale) often outweigh the risks of human oversight—as long as boundaries are designed properly.

Hidden Assumptions

But this thesis rests on several fragile assumptions:

1. Are targets clear enough? Growth "targets" seem simple on the surface (user count, retention rate, revenue), but real growth is often multi-objective (short-term users vs. long-term retention, high-value users vs. broad penetration, brand image vs. rapid growth). When an AI agent makes decisions amid conflicting targets, who verifies that decision aligns with the company's true values?

2. Can boundaries really be drawn cleanly? "Budget safety guardrails" can only prevent overspending, but growth risks extend beyond money. An aggressive user acquisition strategy might bring hordes of low-quality users, damage the brand, or trigger regulatory issues—all outside pure budget boundaries.

3. Will feedback loops accelerate learning or amplify bias? An AI system autonomously optimizing on imperfect data risks getting trapped in local optima. Human review is slow, but it catches systemic blind spots the system can't see. Can the speed advantage of autonomy offset this blindness?

Why Now?

The timing is mature because three things have aligned:

• LLM reasoning capability: Modern LLMs can understand complex business logic, connect insights across multiple data sources, and generate actionable decisions
• Founder knowledge accumulation: The founders have 50+ billion dollars in growth experience and know how boundaries should be set and which decisions can be delegated to machines
• Market pain point: As internet growth dividends fade and traditional growth's labor costs rise, companies urgently need new efficiency leverage

This combination transforms GrowthGPT from "an interesting idea" into "a potentially viable product."