Anant Agarwal, founder of edX, joins to discuss his new role as CEO of Grady, the transformative potential of AI in education, the evolution of online learning, and how AI grading can democratize teaching. He shares insights from his career, the development of AI-powered grading tools, and visions for the future of education.
What Anant Agarwal said
Written from the verified transcript and checked against it. Every figure links to the moment it was said.
Anant Agarwal, former MIT professor and founder of edX, discusses his transition to CEO of Grady, an AI grading startup. He recounts edX's history, including its 2012 launch, 154,000 students in his first course, growth to 100 million learners, and acquisition by 2U for $800 million in 2021. Agarwal explains that Grady solves the long-standing challenge of grading open-response work, using technology from Rutgers and UIC professors that requires no training and handles handwritten work and schematics. He emphasizes that Grady provides instant, detailed feedback at scale, with faculty kept in the loop for review and regrading. He claims AI feedback is preferred by students 100% in AB tests and is 10 to 30 times less error-prone than humans. Looking ahead, he predicts AI will augment teachers, not replace them, with tools like 24/7 tutoring and practice, while teachers focus on mentoring.
Key takeaways
Grady launches as a standalone tool tomorrow and integrates with Canvas, Blackboard, and Open edX.
AI feedback is 10 to 30 times less error-prone than human grading, per AB tests.
Students prefer AI feedback by 100% in AB tests.
Grady can grade 5,000 to 10,000 students in about an hour with detailed comments.
A Georgia Tech AI course on edX has used Grady for a year.
Hello, I am here today with Anant Agarwal, the newly minted CEO of Grady. You may have heard his name before. He is also very well known as the founder of edX. Today we're going to talk about this transition, his first new job in 15 years, and about the state of the market and AI, where MOOCs and higher ed are today versus where they were when he started. But before we do that, I'd love to learn a little more about you. I feel like I know you vicariously because you've been a pillar of the industry for so long. Could you tell us how you got to where you are today and why now is the time to transition into a new organization?
Sure, Matt. First of all, great to chat with you. I was a professor at MIT for nearly 40 years. In 2011, I was the director of the computer science and AI lab at MIT. Alongside MIT and Harvard, we started edX, an online learning platform. We launched edX.org in 2012. I taught the first course on circuits and electronics, and we got 154,000 students from 163 countries in that first course with no marketing. Over the past 15 years, edX has grown to over 100 million learners from 200 countries, with 250 institutional partners and over 5,000 programs. We created innovations like MicroMasters and MicroBachelors. In 2021, edX was acquired by 2U for $800 million. So for the last five years, I've been with 2U. Now we have a similar inflection point with AI. I've been on the board of a startup called Grady, which has solved the AI grading problem. Grady can grade anything—handwritten work, essays, sketches, schematics—with amazing quality and provide instant feedback. This opportunity was incredible, so I decided to come on as CEO while remaining a senior advisor to edX.
Yeah, exactly. The computer science and AI lab has AI as part of its name, so AI has been around for a long time. So at that time, alongside MIT and Harvard, we started edX. It was an online learning platform that enabled learners from anywhere to access free or low-cost learning from top institutions. We launched in 2012, and I taught the first course on circuits. We had 154,000 students from 163 countries in that first course. Today edX has over 100 million learners, 250 institutional partners, and over 5,000 programs. We created MicroMasters, MicroBachelors, and courses that convert to credit. In 2021, edX was acquired by 2U for $800 million. I like to say not bad for a small Cambridge nonprofit. Now AI is transforming everything, and we've launched AI-based learning technologies on edX. At the same time, I joined Grady as CEO because I see an opportunity to democratize teaching and grading. Grading is something professors have to do but don't enjoy, and learners need feedback. Grady can grade anything with high quality and provide instant feedback.
I think there's something important there. I was just on the phone with Mike Olsson at Proctoo, and he was telling me about their origin story, which has similar roots at ASU Online. He built a company around proctoring. The parallel I draw is that you at edX have seen every component of the online degree stack. Here you've decided that grading is either the most important thing to solve or the most eminently solvable by advances in AI. I'd love if you could talk about what made you say, 'This is a problem worth solving independently.'
That's a great question. Back in 2012, we built a machine learning based AI grader at edX. Grading open-response questions and essays was particularly hard. We open-sourced the technology, but it wasn't quite usable—it required training with hundreds of essays and still wasn't there. So grading has remained a challenge. When I saw the technology created by Professors Pericles and Anastasios at Rutgers and University of Illinois in Chicago, I was skeptical at first. I tried it on my own MIT circuits assignments with schematics and sketches, and I was completely blown away. No training needed, and it provided detailed feedback. I said this can really democratize teaching. They are amazing technologists and needed help scaling. I love building things, so I decided to join them to help every teacher become a super teacher. I wrote an op-ed in the New York Times two years ago saying AI can make every teacher a super teacher. This technology can give every teacher the power of 1,000 teaching assistants.
Can you say what it is about what the co-founders have done that makes it work? I get pitched things like converting PDFs into courses or answering homework questions automatically that are just wrappers on ChatGPT. Given your involvement and the other advisers, I assume this is more sophisticated. What makes it more magical and efficacious?
Excellent question. I've tried using ChatGPT for grading, and it sometimes works. The challenge is like pouring crude oil into your car—you need gasoline. LLMs have a sycophancy problem and aren't designed for grading. The founders built a stack of technology that eliminates hallucinations and handles handwritten work and schematics. It has to be right because if even one in a hundred students gets a disastrous result, I have to check everything. The secret sauce is how the stack works together, using multiple engines. Grady is both a standalone tool launching tomorrow and integrated into Canvas, Blackboard, and Open edX. A Georgia Tech AI course on edX has been using it for a year.
9 more exchanges in this transcript
Sign in free to read the rest of this interview. No card required.
Agarwal, A. (2026, June 30). Anant Agarwal, founder of edX, joins Grady as CEO [Interview transcript]. Whiteboard Advisors. CEOInterviews.AI. https://ceointerviews.ai/interview/1055012/
MLA
Anant Agarwal. "Anant Agarwal, founder of edX, joins Grady as CEO." Whiteboard Advisors, 30 Jun. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/1055012/.
BibTeX
@misc{agarwal2026_1055012,
author = {Anant Agarwal},
title = {Anant Agarwal, founder of edX, joins Grady as CEO},
howpublished = {Interview transcript, Whiteboard Advisors. CEOInterviews.AI},
year = {2026},
month = {jun},
url = {https://ceointerviews.ai/interview/1055012/},
note = {Speaker-attributed transcript with timestamps}
}
