Andrew Ng launches LearnVector with $100M from Coursera to build AI tutors that replace classrooms, not teachers

Andrew Ng Launches LearnVector With $100M From Coursera to Build AI Tutors That Replace Classrooms, Not Teachers

Andrew Ng has launched LearnVector, a new AI-native learning company backed by $100 million from Coursera, and his central argument is that the ed-tech industry has been applying AI to learning the wrong way.

Ng co-founded Coursera, founded DeepLearning.AI, and serves as an adjunct faculty member in computer science at Stanford 1. His new company, based in Mountain View, is building what it describes as agentic AI: multi-step systems that reason and act across a sequence of decisions rather than responding to a single prompt 1. The launch page states that chatbots without guardrails harm learning by enabling cognitive offloading, where students finish tasks but fail to develop the underlying skills 1.

Coursera funding its own disruption

Coursera's $100 million investment secures a one-third ownership stake in LearnVector on a fully diluted basis 2. Axios first reported the deal 3.

The investment puts Coursera (NYSE: COUR), a leading global online learning platform, in an unusual position. Its core business delivers one curriculum to many learners simultaneously. LearnVector's stated goal is to replace that model with individualized one-to-one tutoring 1. Coursera CEO Greg Hart said the company believes AI expands the market for learning rather than replacing it, and described the investment as a potential force multiplier that pairs Ng's work in agentic AI with Coursera's content and data from more than 300 million learners 2.

The strategic logic cuts two ways. If LearnVector succeeds in making one-to-one AI tutoring the default, Coursera owns a third of the company that made its existing catalog less central. If LearnVector fails, Coursera has spent $100 million on a company with no product until at least early 2027 2.

What agentic tutoring means in practice

The distinction LearnVector draws between its approach and chatbot tutoring is architectural. A chatbot responds to one input with one output, then waits. An agentic system maintains state across multiple steps, decides what to teach next, adapts to how the learner is progressing, and continues until the user has mastered a skill 1.

Ng frames the history of education on the launch page as a story about economics rather than pedagogy. One-to-one tutoring has always outperformed classroom instruction. The reason it never became the default is that individual tutors are expensive and scarce. AI, in this vision, removes that constraint 1.

Whether multi-step agentic systems will actually produce better learning outcomes than chatbot interfaces is an open question. LearnVector's page invokes research on cognitive offloading but does not cite specific studies 1. The company is hiring AI engineers, learning engineers, and learning scientists to build the system, with products targeted for early 2027 1.

LearnVector also plans to collaborate with Udemy, which combined with Coursera in May 2026 4. No separate Udemy announcement has been made.

The bet underneath the bet

If Ng is right that agentic tutoring is categorically different from prompt-and-answer interfaces, the ed-tech industry's current approach has a structural problem. Companies that have invested in chatbot-style tutoring assistants would not face a feature gap. They would be solving the wrong problem. The question underneath the $100 million is whether the chatbot model is a step toward something better or a dead end that teaches dependence rather than skill. Ng is betting it is the latter.