Brown professors devise course to explore generative AI in computer science education

PROVIDENCE, R.I. [Brown University] - When Anthropic launched Claude Code last year, the AI coding tool set off an earthquake in the computer programming world.

Suddenly, users could simply outline a task in plain English, and an autonomous Claude Code agent could whip up workable code in minutes, sometimes seconds. Soon after Claude Code landed, other companies followed with their own AI coding models, leaving many to wonder what the future for human software developers might look like - not to mention what might become of computer science education.

The advent of Claude Code wasn't a surprise to Kathi Fisler, research professor and director of undergraduate studies for Brown University's Department of Computer Science. But it was a turning point.

"We'd been seeing the whole agentic AI coding thing coming for a couple of years," Fisler said. "It was there, but it wasn't yet really good at what it does. Then suddenly the capabilities of the tools jumped, and we knew we had to teach the students how to work with these things. When trying to get a job or an internship, students are going to be expected to know how to use them."

Not to mention the fact that the tools are now good enough to be dangerous. The code generated by AI agents is far from flawless. At Brown, students often write code for research projects outside the computer science department or for community nonprofit organizations. "If students with limited programming experience are going to use agents when they work with professors on research projects or off-campus partners, now there's an opportunity to do real damage," Fisler said.

To address the issue, Fisler started thinking about how to incorporate agentic programming into the computer science curriculum, working with fellow computer science professors Shriram Krishnamurthi and Michael Littman. Rather than upend the department's entire suite of intro classes, the team opted for a more measured approach that brought students into the process.

They created a class, which ran during the spring semester, called Agentic Studio. The idea was to give a small number of students with at least one prior computer science course a chance to learn firsthand - but in a supervised way - what coding agents can and can't do, where they add value, where they fall apart and what role human software engineers have in working with AI agents.

"We were upfront with the students about the fact that this was experimental," said Littman, who is also Brown's associate provost for AI. "We said: 'We're going to try to figure out how to teach this stuff, and you're going to be our partners in exploring this space.'"

The Agentic Studio course was project-based. Students worked in teams to tackle a selection of software tasks - from creating a "graduation checker" that evaluates whether students have met all of Brown's bachelor's degree requirements for computer science to developing a Signal-like messaging app. Along the way, students kept journals in which they recorded their experiences using AI agents in each project.