Case Study

A front-row view of generative AI at news organizations

By Sonali Verma

October 19, 2026

This article is part of the Lenfest AI Collaborative and Fellowship Program’s 2026 Impact Report.

The Lenfest AI Collaborative and Fellowship Program has given AI practitioners a front‑row view of how generative AI is reshaping work, roles, and culture at news organizations. Our AI Fellows, embedded in newsrooms and product teams, have been catalysts for transformative change over the past couple of years.

When they started in late 2024, OpenAI’s GPT‑4o was the frontier model. Vibe coding and agents were still mostly aspirational, and conversations in journalism focused heavily on copyright and training data.

In the two years since the start of the fellowship, AI has become far more capable, far easier for non‑technical staff to use, and increasingly woven into everyday editorial and business workflows.

How has this evolution affected news organizations – and the work that our AI Fellows do? Here are some key learnings from Fellows Rajesh Barade of The Seattle Times, Mark Chonofsky of Chicago Public Media, and Kevin Hoffman of The Philadelphia Inquirer, shared at a gathering to mark the program’s second anniversary.

Role shift: From engineer to coach

The “AI person,” who was once seen simply as a coder, has increasingly become someone who empowers and enables colleagues to use AI.

“It started off as somebody who could help users automate something that they are doing,” said Rajesh. “Now, it’s shifting more toward a coach-type role, who helps people accelerate their journey, as opposed to somebody sitting in a corner building something.”

Education and training were vital along the way, said Kevin. “We embarked on a teaching journey of, ‘This is what AI looks like, this is how it can help you, this is how you might not want to use it.’”

Culture and trust are still stronger than technology

Trust-building and empathy are prerequisites for creating useful AI tools, each of the Fellows said.

“People are used to getting software systems that work for some startup in Silicon Valley, but don’t really work for them in the news environment for their specific job, and so there’s a lot of just sitting with people and fixing whatever they have,” said Mark.

“It’s just being genuine friends with people. Expressing real interest in what they do goes a long way.”

Kevin, who initially approached the newsroom with caution, found that working with journalists was an eye-opener.

“I quickly realized that I had the best user-testers of all time,” he said. “Journalists are naturally curious and ask questions. So, when I wanted to have people test out things I was building and stress-test different prototypes, it was the exact group who would want to test it. They knew exactly how to break it.”

And if you build something genuinely useful, your journalists can also be your best ambassadors, he said.

“It feels like once we got some buy-in from the newsroom, they bought each other in, and there was a lot of conversation amongst each other: ‘This is actually a really cool tool, I swear. Just give it a shot,’ and suddenly, there magically were more people trying out some of our tools,” Kevin said.

Mark agreed, citing literature that showed “if you want to get one of these interventions to stick, you have to train not one person but two people so they can talk about it. There’s something deep to that.”

AI is easy to work with, but the right user experience matters

The real difficulty does not lie in using AI. It is in integrating AI well into people’s work.

“I think the industry likes to tell itself that doing AI is intrinsically difficult. That it’s very much like computer programming, and I think that the real defining characteristic of these tools is that they are dramatically easier to use than any other computer system that we have ever encountered,” said Mark.

“So the real work is around sitting with users and ensuring that they have the opportunity to do the kind of work that feels intuitive to them.”

Democratization: Everyone can be a builder

Unlike in previous eras of technological change, non-technical staff can now build useful tools themselves. Anyone can simply instruct an AI coding app in their native language to generate code, solving technical problems in real time.

“English is the new programming language,” Rajesh said.

Kevin pointed to the rise of vibe-coding in the Inquirer newsroom. “We’ve really seen it lower the barrier to entry to building some of these more technical tools that might have otherwise been launched by product teams or more technical experts,” he said. In the past, it often took weeks or months before a tool could be deployed.

AI must serve real business and editorial needs

AI is only good when it solves real problems.

“We’re not an AI organization, so we need to have the patience, the time, the judgment, the discernment to figure out whether it solves a business need. If so, we should do it, and if not, we should do something else,” said Mark.

And even as AI proves itself useful across different parts of the news operation, human sensitivity and experience remain irreplaceable, Kevin said.

“At the end of the day, we’re still an institution that relies on human judgment,” he said. “That’s our brand. Whatever we’re building cannot interfere with those sorts of operations.”

Shifting bottlenecks

What is hard and what is easy keep changing as the technology and user attitudes rapidly evolve.

“The bottlenecks are shifting – the hard things become easier, and easier things become harder,” said Rajesh.

For example, initial hesitation from non-technical colleagues around using AI has now given way to vibe-coding, he said.

“When I look at adoption, there’s a shift in perspective. It’s proceeding at a much faster pace. People can totally vibe-code an entire application. That gives them huge confidence.”

The new “hard” thing? Governance, in organizations where everyone has access to AI tools, and many are swiftly building their own applications.

“The level of curiosity has kind of shifted to the other side,” Rajesh said. “I frequently get requests saying, ‘Can we use this connector? Can we connect to this tool?’ We have to think really long and hard. Being able to figure out how to structure the data and the tools that people use is going to be a big thing. I think that’s also going to be a really difficult thing to do, considering the very thin layer of AI understanding across an organization.”