CMU Students Put AI Coding Tools to the Test
Marylee WilliamsTuesday, October 14, 2025Print this page.
It took a room full of Carnegie Mellon University students 13 minutes to build something similar to Instagram. And no one wrote a line of code.
It was the first day and first in-class assignment for students in AI Tools for Software Development, a new experimental course offered by the School of Computer Science's Software and Societal Systems Department (S3D) this fall. But the course goes far beyond recreating social media apps. Students are learning how artificial intelligence tools can be part of the software development process, while professors are determining the best ways to teach computer science while working with these tools.
"We've told the students, you will not write actual code in this class," said Andrew Begel, an S3D associate professor. "Instead, you will tell the large language model (LLM) to do everything for you. If it generates incorrect code, you need to be able to read it and tell the LLM how to fix it, but you won't code the fix yourself."
It's all part of vibe coding. The relatively new term — which has sprung up as people increasingly turn to AI tools for software development — refers to when someone uses an LLM to write code. Begel defines it as "coding with the hope that the output will work flawlessly without testing or training."
Students in AI Tools for Software Development vibe code a lot, but rather than hoping, they're learning how to apply a software engineering mindset to make AI tools more reliable.
Akeil Smith, a junior majoring in business administration, saw the course as essential for his goals and necessary for an interdisciplinary university like CMU. AI tools are here to stay, he said, and courses should reflect that reality. Google's 2025 DORA Report, which surveys trends in software development, found that 90% of respondents had adopted AI tools for tasks such as writing code, analyzing data and more.
"We're at a rare point in history where technology is advancing faster than courses can be created for it," Smith said. "This course is probably the most relevant class at the university in 2025. This technology is impacting the workforce at scale."
On the first day of class, co-instructors Begel and S3D Associate Teaching Professor Austin Henley explained that tech companies expect junior software developers to already have experience using AI tools, like Cursor or Windsurf, when they start jobs.
Both Begel and Henley worked at Microsoft before coming to CMU. Begel researched the human aspects of software, focusing on how software teams collaborate to build large software products. Henley studied AI tools for software developers and published papers about how novices use LLM code-generators and ways to improve AI-assisted code editors.
Henley said he envisioned teaching a class like this one when he joined the SCS faculty.
"I thought up a course like this when I saw the push internally at Microsoft to use AI products for our own day-to-day work," Henley said. "We were using AI to build AI."
Miguel Almeida, a master's student in the Languages Technologies Institute, appreciates that SCS is encouraging students to learn how to better adopt AI tools.
"I hope to learn how to integrate AI tools into my work, no matter what new features are added to these applications," Almeida said. "AI developer tools are moving at a thrilling pace, so I look forward to learning how I can leverage them to create value at a very basic level."
The class is broken up into lectures, in-class activities and projects. Mob programming, a type of chaotic group programming activity, is a staple of the in-class activities and involves the class calling out commands that a student standing in front of the room types into an open AI tool. The students have to collaborate to help the AI tool achieve a goal based on something they learned recently in lecture.
"As the mobbing activity proceeds throughout the semester, they're going to get better at knowing how to convert requests from classmates into commands that they should give to the LLM," Begel said.
This class is also an experiment for S3D. Michael Hilton, a teaching professor and S3D's associate department head for education, said this is a chance for faculty to see how AI developer tools can be taught effectively. For him, Henley and Begel are the perfect instructors to approach this experiment because of their research experience with these tools.
"It's a bit unusual to pitch a class and essentially say, 'We don't really know what we're doing, and we're going to figure it out as we go along,'" Hilton said. "But I think we're at a unique moment in time and this is one of the only ways to approach it. If we wait 10 years until everybody has figured out everything on how to teach AI and software development, it's going to be too late for a lot of people."
At the beginning of December, the students will present their final projects — web applications created with the help of AI. Almeida's group plans to make an app that draws inspiration from Slack, but it will take much longer than 13 minutes to create because his group isn't just vibe-coding. They're also using newly learned software development processes to make the app. But they'll still do it without writing a line of code.
Aaron Aupperlee | 412-268-9068 | aaupperlee@cmu.edu