PROVIDENCE, R.I. [Brown University] — When Anthropic launched Claude Code final yr, the AI coding instrument set off an earthquake within the pc programming world.
Instantly, customers might merely define a activity in plain English, and an autonomous Claude Code agent might whip up workable code in minutes, typically seconds. Quickly after Claude Code landed, different firms adopted with their very own AI coding fashions, leaving many to marvel what the long run for human software program builders would possibly appear to be — to not point out what would possibly grow to be of pc science schooling.
The arrival of Claude Code wasn’t a shock to Kathi Fisler, analysis professor and director of undergraduate research for Brown College’s Division of Pc Science. Nevertheless it was a turning level.
“We would been seeing the entire agentic AI coding factor coming for a few years,” Fisler mentioned. “It was there, however it wasn’t but actually good at what it does. Then abruptly the capabilities of the instruments jumped, and we knew we needed to train the scholars learn how to work with these items. When attempting to get a job or an internship, college students are going to be anticipated to know learn how to use them.”
To not point out the truth that the instruments at the moment are ok to be harmful. The code generated by AI brokers is way from flawless. At Brown, college students usually write code for analysis initiatives exterior the pc science division or for group nonprofit organizations. “If college students with restricted programming expertise are going to make use of brokers once they work with professors on analysis initiatives or off-campus companions, now there’s a possibility to do actual injury,” Fisler mentioned.
To deal with the problem, Fisler began desirous about learn how to incorporate agentic programming into the pc science curriculum, working with fellow pc science professors Shriram Krishnamurthi and Michael Littman. Fairly than upend the division’s whole suite of intro lessons, the staff opted for a extra measured method that introduced college students into the method.
They created a category, which ran throughout the spring semester, referred to as Agentic Studio. The concept was to offer a small variety of college students with not less than one prior pc science course an opportunity to study firsthand — however in a supervised manner — what coding brokers can and may’t do, the place they add worth, the place they collapse and what position human software program engineers have in working with AI brokers.
“We had been upfront with the scholars about the truth that this was experimental,” mentioned Littman, who can be Brown’s affiliate provost for AI. “We mentioned: ‘We’ll attempt to determine learn how to train these items, and you are going to be our companions in exploring this area.’”
A classroom exploration
The Agentic Studio course was project-based. College students labored in groups to sort out a choice of software program duties — from making a “commencement checker” that evaluates whether or not college students have met all of Brown’s bachelor’s diploma necessities for pc science to creating a Sign-like messaging app. Alongside the best way, college students stored journals during which they recorded their experiences utilizing AI brokers in every undertaking.