en.Wedoany.com Reported - Yimao Zhou, the 23-year-old founder of Emagen AI, has launched an "OS Level Agent" called Cagen, arguing that the current AI agent industry is optimizing the wrong unit. Zhou criticized that existing products only make individuals more efficient while exacerbating team collaboration costs. The company, backed by MiraclePlus (formerly YC China) and its founder Qi Lu, is betting on building an operating system where AI drives work and proactively engages humans, rather than the other way around.
New AI agent startups are popping up every week, capable of writing code, drafting emails, generating slides, and analyzing data. Yimao Zhou believes these products are solving the wrong problem. He points out that knowledge workers spend 60% of their time on collaboration costs, including syncing progress, writing status updates, and passing information. AI tools mainly optimize the other 40% of "doing" tasks while completely ignoring collaboration issues, which actually makes teams worse.
Cagen is described as an "OS-level agent." Zhou explains that existing SaaS companies like Notion, GitHub, and Salesforce are embedding AI, but their incentive is to make their own products stickier rather than connecting across tools. Cagen, as an independent layer, can solve this structural problem. Unlike AI products that wait for human queries, Cagen's agents have goals and context, continuously reasoning based on team objectives and proactively engaging humans when needed. Humans are treated as resources within the system, not operators of the system.
Zhou says that after six months of using Cagen, the organizational intelligence accumulated by the team—including decision-making patterns, communication habits, quality standards, and workflow knowledge—becomes its unique moat. This organizational memory has network effects, where one person's learning benefits everyone and every agent in the team. He believes that personal AI makes one person better, while an OS-level agent makes the entire organization smarter as a unit.
Cagen's early customers are not tech startups. Zhou reveals that the product is currently deployed in a boutique hotel in Pittsburgh, which has high operational complexity, handling guest communication, maintenance coordination, scheduling, supplier management, and numerous cross-role handoffs. He calls this the best entry point because the hospitality industry is one of the most operationally dense environments and is not deeply tied to a specific tool ecosystem.
Regarding cross-industry expansion, Zhou says Cagen's roadmap is to internalize the consulting process into the product. Ideally, users only need to describe their team's daily work and company goals, and the system automatically understands and builds the correct workflows. This process aims to avoid becoming a custom consulting firm and instead solve a scalable platform problem.
Qi Lu decided to invest after just 10 minutes of a 30-minute demo. Zhou believes Lu invested based on his judgment of the industry landscape. Lu's career has been at the operating system level, and he recognized the distinction that "the entire industry is building at the wrong layer." Zhou says Lu and MiraclePlus supported this judgment.
Facing strong competitors like Claude Code, OpenAI Codex, and OpenClaw, Zhou believes they are not competing because they are playing different games. He points out that these products enhance individuals, but no one is building an operating system. Data shows that Claude Code's annualized revenue exceeded $2.5 billion by early 2026, contributing to Anthropic's mid-year $44 billion total run rate; OpenAI Codex has 5 million weekly active users; and OpenClaw has garnered over 370,000 stars on GitHub. Zhou says Claude Code's "Agent Teams" feature only allows one person to orchestrate multiple AI instances without shared context or organizational memory. He states that the infrastructure for AI-assisted coding (coding harness) is largely solved, while the infrastructure for AI-assisted work—coordinating teams, managing goals, orchestrating humans and agents—remains a vast blue ocean.
Looking to the industry's future, Zhou predicts that most of today's AI agent startups will fail. The survivors will be those building at layers that are not easily absorbed by foundation models. For Emagen AI, this means the OS layer. He believes that as AI operating systems mature, the minimum viable team size for a serious enterprise will collapse—what once required 50 people will only need 5 plus an AI operating system.
Yimao Zhou is the founder and CEO of Emagen AI, the company behind Cagen. He studied medicine as well as cognitive philosophy and philosophy of science at Shanghai Jiao Tong University, and is the youngest founder in MiraclePlus's F24 batch. For more information, visit cagen.ai.
