The School of Lee

Lee Robinson is the archetypal developer educator. Can he out-teach AI?

Technology, if you haven’t heard, makes people anxious. It makes them especially anxious when it threatens their jobs — or at least, when they think it does. In the not too distant past, “learn to code” was a much-panned piece of advice offered by journalists to coal miners whose jobs were in decline as a result of technological and economic shifts. In recent years, the coal miners have stabilized, and a number of people have turned the table on the journalists, suggesting they “learn to code.” What few expected is that “learn to code” would itself become an ironic piece of advice, given the dazzling progress in AI coding specifically. I guess we’re all going to have to learn to mine!

Alright, I’m kidding. For as long as technology has been changing work, education has been important to help individuals and societies adapt. It’s still true. Learning about code and computers is very worthwhile, and it’s tremendous fun. AI and coding may present a unique conundrum because in the history of education, the subject you’re learning has not been able to teach itself. “Chemistry” can’t teach chemistry. But “AI” can most certainly teach AI. AI can teach you to code, because it knows how to code, which is begging the question... why bother at all?

It’s a question that few people have thought more about than Lee Robinson — “leerob” for short on GitHub and X. Robinson grew to be a well-known figure among developers as one of the main public faces of Vercel, the company that develops the open-source Next.js framework for web applications; today, he’s VP of Developer Education at Cursor, the $30-billion company whose eponymous coding tool has enjoyed one of the most impressive takeoffs of any AI product since the launch of ChatGPT.

I met Robinson in San Francisco on one of his regular trips from Iowa, where he lives (and where I grew up, too). He flies in every few weeks to spend time at Cursor’s offices, of which there are now several in San Francisco’s North Beach neighborhood. He spends a few days, then flies home. It’s a rhythm he settled into during his five years working at Vercel, and one that tells you a lot about how he thinks. “When I’m here in San Francisco, there’s this amazing buzz in the office and energy with things happening. I love it for a week, and then by the end of the week, I feel like I can get work done better when I’m back in my cave in Iowa.”

“I used to have an office upstairs,” he told me as we walked around North Beach. “Then we had a daughter and that was not gonna fly. So I moved my office into the basement. It has a closed door, and I invested probably more than any reasonable person would in having an immaculate camera setup. I plug in a couple things, turn on a couple lights, and I’ve got studio quality video in my home.”

✺

Robinson grew up in small-town Iowa, went to Iowa State University for computer engineering, and stayed for his first jobs as a software engineer—including at Workiva, a publicly traded software company based in Ames (where Iowa State University is), and Hy-Vee, an Iowa-based grocery that is the dominant grocer in Iowa and the leading brand in Nebraska, South Dakota, and Minnesota, too. One of his last projects as an engineer at Hy-Vee was standing up a grocery delivery platform in 2020.

Robinson had nearly left Iowa in 2018. He had a job offer to be a product manager at a startup in San Francisco. In the end he didn’t take it. At the same time, he began to appreciate teaching people about code in his free time — helping people “to understand why we’re doing all this stuff in the first place. It’s not just about writing a bunch of funny characters into a computer, but it’s about building something great.”

Yet he noticed a particular content gap for the type of web development work he was doing with React (a Javascript framework maintained by Meta) and Next.js.

So, he made the content himself, first short video tutorials and blog posts, and eventually paid courses. Because Next.js was still young and almost nobody else was producing material about it, he became a de facto ambassador. “It wasn’t really intentional,” he told me, “but I kind of became the sole result when you Googled Next.js. Basically the only results back in the day were my videos.” The first course that got traction had a hundred paying customers, and the feeling stuck with him. “At the time, that was insane to me.”

The team at Vercel — just a few dozen people at that time — noticed and asked if he’d ever considered doing it full-time. “I mean, actually, no, I haven’t ever thought about doing it full time,” he recalled thinking. “I’ve just been kind of doing this on the side. What does that even look like?” The answer, it turned out, was a lot of different things. And over five years, he helped define a role that is now in high-demand at the most important technology companies engaged in a brutal contest for the love and attention of developers and users. And he managed to do it while staying put in Des Moines.

Vercel is a fairly unique model in tech: a venture-backed company built atop an open-source framework that anyone can use for free, including some of the largest businesses in the world. The tension is obvious, because the business can only thrive when people pay for what are essentially optional services (Vercel has hundreds of millions in revenue from its hosting services).

“It’s funny when you watch people try to copy Vercel’s business model,” Robinson told me. “It’s kind of an insane thing to do because it’s incredibly hard to do well, to have both a successful open source framework and community and then build a successful venture scale business on top of it—and make them not conflict with each other.”

One gets the sense that maybe what other businesses lacked but which Vercel had was… someone like Lee Robinson, a figure whose personality and presentation exude trustworthiness. “I was not doing it just for the love of the game. I wanted the company to succeed,” he told me. And the success of Vercel, to Robinson, was proof enough that you could succeed on both dimensions.

This article appears in print in Arena Issue 007! To get Arena delivered to your door, subscribe at arenamag.com

The release of Anthropic’s Claude Sonnet 4 model in May 2025 represented a turning point in Robinson’s thinking about what DevRel was going to mean, and what role he was going to play. “Now is the time to change,” he recalled thinking. “Otherwise, I don’t know what developer relations or my job will look like in a couple of years. I can keep talking to them about Next.js and React. I’ve been doing that forever. But the AI models are getting very good at that. So what do they actually need help with?” He answered his own question: “They’re going to need to figure out how to navigate this strange, weird AI journey in a way that’s not too hypey but not pessimistic either.”

He joined Cursor in July 2025, when the company had around 80 employees. Its workforce has since grown to several hundred.

“There’s a lot of marketing around AI that developers are allergic to,” Robinson said. “They see the press releases and the launch videos and the product demos, and they just don’t trust it.” He said he wanted to do things a little differently.

“My goal in joining Cursor was to try to give people a pragmatic, cautiously optimistic view of this new world and to bring along these millions of developers who hear these news headlines. Developers are worried about their jobs; they’re worried about the future of this industry; I want to show them there is a path to prosperity where we will be able to do incredible things.”

In the same way that “Uber for X” once described an entire generation of startups — not always positively — “Cursor for X” has become a stand-in for the type of AI tools that makes a complex domain feel suddenly accessible. Cursor for law. Cursor for accounting. Cursor for architecture. Cursor for marketing. The name has become synonymous with the idea that a professional tool can be powerful and approachable at the same time. For someone whose career for the last few years has been about making complicated technology feel accessible, Cursor is a great perch for Robinson, and it keeps him connected to both millions of developers and some of AI’s top luminaries.

“I’ve had CEOs reach out who haven’t been coding for ten years who say, ’thank you for explaining this in a way that was accessible and digestible for me.’ And that’s my job: to cut through the noise and try to explain things in a way that a beginner would understand but that an expert would also find helpful.”

Robinson pointed out to me that in 2025, the top model on SWE-Bench — a benchmarking test that grades AI systems on their ability to solve real software engineering problems — changed more than a dozen times. “I think it’s great,” Robinson said, “because with so much competition, the best product truly wins.” And building the best products for coding is Cursor’s focus (the company develops its own language models, but all major models are available in Cursor’s products).

Robinson traces three major shifts in AI-assisted coding. The first was autocomplete: AI finishes your line of code, not unlike lane assist in a car. GitHub Copilot pioneered this, and Cursor followed with its own model, Cursor Tab. Developers were still writing code mostly by hand, but the AI could complete a line. The second shift was putting a chatbot inside the editor, replacing the need to search Google or Stack Overflow. The third shift was autonomous coding agents. “When you are working with an agent that is autonomously taking actions for you, you’re trying something that’s closer to self-driving, where the car is mostly driving itself. You’re still giving feedback and you’re nudging the wheel, but primarily you’re charting the course for where the car goes.”

Early agents could modify code inside a single file for tightly defined tasks like renaming a function, or moving a block of code around. As the models improved, they could work across multiple files, run terminal commands, and handle the kind of arcane operations that developers used to have to memorize. “Running shell commands, using a terminal — these are things that developers had to do by hand and remember all of these magic combinations of characters,” Robinson said. “But now, AI models can remember all of that stuff for you.”

Continuing the metaphor a bit, Robinson compares the new generation of coding agents to a fleet of cars unleashed on an entire city and allowed to drive freely: “We’re starting to reach the point of full self-driving for agents, where they can run for hours or weeks across entire code bases with millions of lines of code.”

When I asked whether looking at code would remain the primary way software engineers build software five years from now, Robinson didn’t hesitate. “Probably not. You would probably be looking at something that is much more like English. There will still be code. You will still have to think about the code. You might have to debug the code and dig in. But the primary interface will look more like English if the current trajectory continues to play out.”

And the ways in which people can interact with code using Cursor are growing in number, too. “If you want to make changes to your website from your phone, in English, Cursor has an app for that. You can tag Cursor in Slack and instruct agents to go fix bugs that users notice.”

Robinson himself welcomes this new paradigm, even though he isn’t exactly sure where it’s headed. “As an engineer, I loved code because of what it helped me build. I was never in love with looking at the code. But there are people who love coding for the artisanal, handcrafted code, and I think they’re having a bit more of an existential crisis right now.”

✺

Talking to Robinson, one notices that like most great teachers, he has a habit of explaining his opinions by recounting the questions he asks himself. He can marshal history and metaphors in his explanations. And it’s very clear that part of why he enjoys teaching so much is that he enjoys learning, including — or especially — when it’s a slog. “AI models have this funny way of tricking you into thinking that you learned something when you didn’t really actually learn something,” he said. “It’s like this short term gratification of seeing that it works, without the long struggle in the middle to make it work.”

Robinson doesn’t think there’s a shortcut around that struggle, and he doesn’t think there should be. “Learning anything difficult is a struggle. Like, it’s gonna suck. But part of accomplishing and learning really hard things is the grind not understanding, then trying and figuring it out. If you rely on AI too much and you outsource the thinking, you don’t really have a great foundational knowledge for how it works.”

When he said that, it reminded me of one of the first things he’d said as we began our walk from Cursor’s original office down toward the San Francisco waterfront — that he fell in love with teaching code in part to remind people that engineering isn’t an end itself, but a tool to build “something great.” And that, in a single idea, is the school of Lee. So, build to your heart’s content. Make Cursor for Architecture, or Cursor for Boats. Airbnb for Dogs, or Uber for Cats are options — the age of AI has made it exceptionally easy to build these new things — but the ease of it all presents some really hard choices. Life wants you to figure out what you actually care about, what you think “great” means. So, the one thing you might want to hold off on is “Cursor for Thinking,” because that’s worth doing the hard way.

At the very end of our walk, Robinson gave me this thought: “I always thought that at some point I’d have to move here to San Francisco, that it’d be the only way I could achieve the success that I wanted. And surprisingly, that hasn’t been the case.”

When Robinson started working at Vercel in 2020, San Francisco was in something of a nadir. Companies were openly contemplating leaving. A number of prominent individuals did. Staying in Iowa at that moment was easy to explain — why would you move toward the chaos? But in 2026 as a major figure at one of AI’s top companies, Robinson’s basement cave in Iowa is a far more significant choice. San Francisco is resurgent, the most important nexus of talent and capital in the history of the technology industry.

But for him, it is a considered choice, and I can say with confidence he didn’t outsource his thinking to anybody else. “I’ve spent more time than I’d like to admit in the Denver airport on layovers,” he lamented to me in a way that I understand very well as a fellow Iowan, having taken direct flights to my hometown only a handful of times ever (and certainly not from San Francisco).

“Iowa is not a travel destination, it’s not a luxury resort. But it’s a great place for me to raise my family, to be close to my parents and to my wife’s parents. It’s a slower pace of life and I do appreciate that a lot. But I also love San Francisco. I love coming here and being involved in the thick of building the future and the incredible talent density. I feel very fortunate that I can do both.”

Maxwell Meyer is the founder and Editor of Arena Magazine, and President of the Intergalactic Media Corporation of America. He is on X @mualphaxi