Image
Loading...
The trophy for “most successful failure” was given at the end of the semester.
In a class focused on artificial intelligence, it stood out. Students had spent months building tools for real newsrooms, testing ideas, and revising them when they fell short. Some projects worked. Others stalled. One team earned recognition for a project that showed ambition and effort, even if it never fully came together.
That was the point.
In the course known as the AI Journalism Lab at the Sanford School of Public Policy, professors Bill Adair and Tyler Dukes ask students to approach a fast-moving, often uncertain technology with curiosity and a willingness to experiment. The class is structured around a simple idea: build something useful, test it in the real world, and learn from what happens next.
“Failure IS an option,” the course syllabus tells students early on. “Some tools will work; others won’t. But you’ll learn from both.”
Across the semester, students partnered with local journalists to identify time-consuming tasks and build tools to help. Their projects ranged from tracking grocery prices to organizing public records, all grounded in the daily work of reporting.
The students came from a wide range of backgrounds, including public policy, computer science, engineering, and journalism. Some had newsroom experience. Others had never worked in journalism before. Together, they brought different perspectives to the same problems.
The result is a classroom that feels less like a traditional lecture and more like a workshop. Students move between coding, client meetings, and presentations, sharing progress and setbacks along the way.
“They’re doing the thing,” Adair said during one class discussion. “There aren’t enough classes like that.”
The question is not whether journalism should use AI, but how. The goal was to build tools to support journalists, not replace them.
Students Ishan Vyas, Lucy Glynn, and Mekhi Patterson ("Newsletter Aggregator" presentation).
Building tools for the newsroom
Many of the projects began with the same question: What takes up too much time in a journalist’s day?
Image
For editors at The 9th Street Journal, the answer was the inbox.
Each day, managing editor Alison Jones receives newsletters from local outlets, government agencies, and community organizations. Keeping up with them is essential, but it is time-consuming.
Students Ishan Vyas, Lucy Glynn, and Mekhi Patterson built a newsletter aggregator to help.
The tool collects emails from a dedicated inbox, pulls out the most relevant stories, and delivers a single digest each morning. It groups coverage by topic and includes links back to the original reporting.
“The question is not whether journalism should use AI, but how,” the team wrote in their project manual.
In their design, AI plays a narrow role.
“The system uses AI selectively,” the manual explains.
Image
From public records to grocery prices
Other groups tackled different parts of the reporting process.
Valentina Garbelotto, Gemma Tutton, and Fiona Loughran worked with WRAL investigative reporter Sarah Krueger, who regularly reviews public records to identify new leads.
“She was checking public records every morning to see if there was anything interesting,” Garbelotto said. “But it was taking her a really long time.”
The team built a tool that scrapes recent public records and sends a daily summary, highlighting patterns and linking directly to original records so reporters can verify the information themselves.
A little AI magic saves time on warrant reading
Image
Another group focused on search warrants, which Krueger also reviews as part of her reporting.
Reece MacKinney, Amalie Seth, and Gracie Abernethy developed “Warrant Wizard,” a tool that extracts key details from dense legal documents and organizes them into a spreadsheet.
“The whole system is designed to fast-forward the reading process,” MacKinney said during the presentation, “without changing the trust that this system is.”
The team tested the tool on dozens of warrants and focused on accuracy above all.
“Not a single one had hallucinations,” Abernethy said.
When the system could not find information, it left fields blank.
“It’s better to leave something blank than to make something up,” she added.
Working through the limits of AI
Image
Not every project relied heavily on artificial intelligence. In some cases, students moved away from it.
Noor Nazir, Sophia Hinshaw, and Max Tendler worked on a grocery price tracker. The group started by using AI to collect pricing data across stores but ran into problems with accuracy.
“If you’re checking every single price, functionally that does not save you any time,” one student explained during the presentation. “So we wanted to create a system that was actually getting the right prices.”
Instead, the group built a tool that pulls data directly from store APIs, allowing users to compare the cost of a standard basket of goods across locations.
Automating the routine, preserving the judgment
For Indy Week, the challenge was assembling a weekly events calendar.
Image
Editors spent hours visiting dozens of websites, copying event details, and formatting them into a single document.
“This takes eight man-hours every two weeks,” Brenton Wang said. “That is eight hours you are not doing journalism.”
Their tool automates much of that process by pulling event data into a shared spreadsheet, where editors can review, edit, and export listings.
The system still requires human input.
“The system is not fully automatic and still requires human review,” the project manual explains. “But it dramatically reduces the amount of time spent collecting and formatting event listings.”
Image
A different way to approach AI
Across projects, a pattern took shape.
Students used artificial intelligence in focused ways. They limited what it could do and where it could pull information. In some cases, they removed it from parts of the workflow entirely.
“The goal was to build tools to support journalists, not replace them,” the newsletter team wrote.
That mindset shaped the course itself.
Rather than treating AI as something to avoid or adopt wholesale, the AI Journalism Lab gives students space to experiment. They test ideas, encounter failures, and refine their work with feedback from working journalists.
By the end of the semester, each team presents a product. Some are ready for use. Others need more development. All reflect the same process.
“Failure IS an option,” the syllabus reminds them.