Exclusive: NetEase’s Youdao CEO Explains How AI Agents Could Build a Future of Virtual Teachers

The application of artificial intelligence (AI) in education is evolving rapidly, moving beyond simple chatbots and automated content creation towards more advanced “agents” — autonomous systems capable of handling complex educational tasks. NetEase Youdao, a major Chinese online learning company, is at the forefront of this trend, focusing on the development of AI agents to revolutionize learning experiences[para. 1]. According to Zhou Feng, CEO of Youdao, AI is increasingly becoming like an operating system, with agents required to build sophisticated educational applications on top of it[para. 2].

Since early 2025, the education sector has seen an influx of interactive “AI teachers” embedded into courses and tools. These AI agents can break down complex problems, interact more intelligently with students, and demonstrate clear commercial value. As Zhou pointed out, 2025 has emerged as a pivotal year for real-world AI applications, particularly in business and government, with mid-sized models and vertical applications offering significant growth potential, though the rise of agents has been faster than anticipated[para. 3][para. 4].

Three major themes have emerged in the industry for the year: the deployment of AI applications, advancements in model capabilities, and the progression of AI agents. Investment in chatbots, though once dominant, has waned, giving way to a surge in programming products, partially inspired by the monetization of similar technologies by firms like Anthropic[para. 5].

On August 20, Youdao launched the “Space X” AI answer pen, a new device capable of generating blackboard-style explainer videos in response to student queries. This process involves multiple steps — from problem comprehension to scriptwriting and audiovisual production — and requires significant computational resources due to the complexity of code generation[para. 6]. Despite high token consumption, the outcome is more engaging explanations tailored to students and appreciated by parents[para. 7].

Zhou likens AI advancement in education to the five levels of autonomous driving: Level 1 signifies basic Q&A, Level 3 involves active learning support (where the answer pen currently sits), and Level 5 embodies research and innovation. While most educational AI tools are currently between Levels 3 and 4, the next breakthrough is seen as the emergence of the “virtual teacher” with persistent memory and strategic teaching ability[para. 8][para. 9][para. 10].

A true virtual teacher would integrate multimodal abilities, long-term memory, interactive capabilities, data management, pedagogy, and hardware to ensure students not only receive content but actually learn and progress[para. 11][para. 12]. Despite the value of extensive data and educational content, the critical challenge remains in applying these tools to real-world learning problems, rather than focusing solely on model size or raw computing power[para. 13].

Currently, only AI chatbots and programming tools have scaled globally, with OpenAI’s ChatGPT expected to generate more than $10 billion and Anthropic’s coding innovations also earning billions. While education applications are not yet at this scale, Zhou is optimistic that they will get there as viable scenarios emerge[para. 14][para. 15].

Youdao’s AI offerings span both software and hardware, including virtual language teachers, math problem solvers, plagiarism detectors, and AI answer pens like Space One and Space X. While these solutions are currently at Level 3, Zhou expects Level 4 capabilities within a couple of years, corresponding with a projected mobile internet-style growth cycle for generative AI[para. 16][para. 17].

Chinese K-12 education remains largely paper-based, necessitating seamless integration of AI tools with traditional study habits. Zhou emphasizes that AI interventions should support, not replace, student effort, positioning answer tools as aids when learners are genuinely stuck, especially in subjects like math where recent AI advances have brought performance close to the human level[para. 18][para. 19][para. 20].

Achieving high accuracy (96% for K-12 problem-solving) required years of work, particularly in math and physics. Continued improvements involve better image recognition and formula parsing. Unlike competitors, Youdao develops proprietary models, focusing on speed and usability — their smaller models are faster and more cost-effective than large, cloud-based ones, making them suitable for consumer hardware[para. 21][para. 22][para. 23].

Declining inference costs and rapid hardware advancements have made devices like the Space X pen more practical, affordable, and widely adopted, especially among high school students, who use them frequently for science and math queries. Zhou frames AI answer pens as productivity tools, highlighting their ability to dramatically reduce problem-solving time, thereby delivering tangible efficiency gains in education[para. 24][para. 25][para. 26][para. 27].