Andrew Ng launched LearnVector on July 28 with a $100 million strategic investment from Coursera, returning to the problem that produced his first education company: how to give more people access to effective teaching.
LearnVector is building an AI learning guide that will plan a path for each user, adapt its instruction and keep working with the learner until they master a skill. The Mountain View, California-based LearnVector plans to show its first products in early 2027. LearnVector is already recruiting AI engineers, learning scientists and other employees to work on-site.
Coursera's backing values an unshipped product and its founding team at a substantial price. Reuters reported that Coursera will receive one-third of LearnVector, implying a valuation of about $300 million. Coursera and LearnVector have publicly described the transaction as a strategic investment rather than a conventional venture round.
The size reflects the person Coursera is backing. Ng co-founded Coursera in 2012 after teaching a Stanford machine-learning class online to more than 100,000 students. He previously led Google Brain, served as Baidu's chief scientist and later founded DeepLearning.AI, extending a career spent moving AI research and education toward wider use.
LearnVector gives Ng another attempt at the part of online education that prerecorded courses never solved. Coursera expanded where a student could learn, he wrote on LearnVector's site, while instruction remained "one-size-fits-all." His new bet is that AI agents can change the lesson itself for each learner.
A founder-backed sequel to Coursera
Coursera is supplying LearnVector with capital, content and a potential route to market. Ng said LearnVector plans to work closely with Coursera and Udemy, which completed their combination in May. Coursera says the combined platform reaches more than 290 million learners and 18,000 enterprise customers.
The arrangement keeps Ng tied closely to Coursera. He is founder and CEO of LearnVector, in which Coursera is taking a large minority position, and remains the Coursera co-founder most associated with its original education thesis.
Coursera is making a meaningful balance-sheet commitment. Coursera CEO Greg Hart said on LearnVector's site that the investment could be a "force multiplier" for growth by pairing Ng's work on agentic AI with Coursera's platform and assets.
That structure gives LearnVector distribution before it has a finished product. It also gives Coursera an outside product lab led by the founder most closely associated with its original education thesis. Coursera can retain its existing catalog and credential business while gaining an ownership stake in Ng's attempt to rebuild the learning experience around AI agents.
The product remains under definition
LearnVector's public description is specific about behavior and early about implementation. The planned guide will understand a learner's goals, prepare a path, deliver instruction step by step and measure whether skills are retained. LearnVector says the experience should resist the tendency of general-purpose chatbots to provide an answer before the user has learned how to produce it.
Ng told Reuters that LearnVector is still defining its course offering, user interface and pricing. LearnVector plans to use third-party foundation models rather than train a base model from scratch, according to the report. Ng also said LearnVector expects to sell courses to companies, governments and higher-education institutions.
That commercial plan starts narrower than LearnVector's broad mission of building a personal learning experience for everyone. Institutional customers offer established budgets and groups of workers whose skills can be assessed against job requirements. They also give LearnVector a clearer way to measure whether its agents improve workplace capability, rather than counting conversations or completed lessons.
The distinction between task completion and skill acquisition sits at the center of Ng's pitch. LearnVector cites a field experiment involving nearly 1,000 high school math students. Students given access to an unrestricted GPT-4 interface performed better while using it, then scored 17% worse than a control group after access was removed. A tutor configured with educational safeguards largely mitigated that decline.
LearnVector is using that result to argue for product constraints as a feature. An effective tutor must decide when to explain, when to offer a hint and when to make the learner continue working. The harder engineering problem is measuring mastery over time and preventing the underlying model from reverting to the fastest answer.
Distribution will not settle the product question
LearnVector enters a market where large AI platforms already recognize that answer engines can undermine learning. OpenAI introduced ChatGPT Study Mode in July 2025 with guided questions, knowledge checks and step-by-step responses. Google's Learn Your Way adapts source materials to a student's grade level and interests while offering quizzes, audio lessons and other formats.
Ng's advantage is the combination of his education record and Coursera's distribution. LearnVector can also ground lessons in an existing body of instructor-produced material rather than asking a model to assemble a curriculum from its own generated responses. Coursera and LearnVector still have to prove that this content can be converted into adaptive instruction without weakening its accuracy or educational value.
LearnVector says it expects to have products to show by early 2027. Until then, Coursera's $100 million is a bet on Ng's ability to solve the instructional problem left behind by the online-course industry he helped create. Ng spent the last 15 years helping make courses available at global scale. LearnVector is his attempt to make the teaching inside them personal.