Case Study

Lenfest AI Fellow spotlight: Aaron Brezel

ProPublica

October 19, 2026

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

As an AI Engineering Fellow at ProPublica, Aaron Brezel is designing human‑centered AI tools that help investigative reporters work more efficiently, without sacrificing rigor or agency. We spoke with Aaron about what he’s learned so far, what the industry gets wrong about AI, and why he believes human judgment will matter more than ever in the years ahead.

About his role and the problem he is trying to solve

Aaron is one of two AI engineering fellows at ProPublica, alongside his colleague Dana Chiueh. His work is squarely focused on the investigative process, rather than on audience- or business‑side tooling.

Aaron’s mandate is to work in the newsroom and support reporters directly in producing journalism. With a master’s degree in journalism and in engineering from Columbia University, he draws heavily on his prior experience at The Washington Post, where he worked on the Reporting Tools team.

“We were a group of engineers with engineering backgrounds,” he said. “Our remit was to collaborate within the newsroom and understand the common problems that they were experiencing and build durable systems to support them in the future.”

At ProPublica, he has taken the same approach: embed, observe, collaborate, and then build.

Reporters come to him with “AI‑shaped problems” – massive document sets, messy data, and opaque systems, for example. He helps them break those challenges down into workable processes.

Over time, he’s identified repeating patterns and is now building at least one core product to support the most common use cases – an open‑source Google Sheets sidebar that lets reporters run structured, row‑by‑row AI operations using Gemini and related tools for tasks like document OCR and analysis, directly inside the spreadsheet environment that they already work in.

What has surprised him most during the fellowship

“What surprised me most has been the eagerness that reporters have displayed towards using AI tools,” Aaron said. “There’s a lot of general consternation about AI, but I think it’s actually directed more towards AI mediation between a reporter and their reader. There’s a lot less anxiety around reporters using the tool for their own purposes to conduct investigations.”

In other words, reporters are wary of AI as a publisher, distributor or gatekeeper, but enthusiastic about AI as a backstage assistant that helps them uncover stories faster and more deeply.

“I get more requests than I can deal with, which is a good thing as far as I’m concerned,” he said.

“Our remit was to collaborate within the newsroom and understand the common problems that they were experiencing and build durable systems to support them in the future.”

What the industry gets wrong about AI

For Aaron, the biggest misconception is treating “AI” as if it were a solution in itself.

“The thing that people get wrong is that AI in itself is not a solution to anything. It’s a model. It takes an input and spits out output.”

He argues that the fundamentals of good product thinking and reporting support haven’t changed: Understand real user problems, iterate through solutions, and focus on the minimum valuable product to build.

What has changed is the speed and volatility of the environment

“We just have to do the same things that we’ve always been doing. Unfortunately, now we just have to do it 10 times faster because everything is moving so quickly,” he pointed out.

He’s particularly critical of AI products that over‑promise “magic” and encourage reporters to hand over too much of their process. “AI is a means to an end. You really have to understand where and how you’re actually applying AI.”

One lesson from his work that other newsrooms could apply tomorrow

Build direct, human relationships between technical staff and reporters, and treat the job as enabling reporting, not just shipping software.

“I am a big proponent of direct relationships with the people you’re trying to build for. There’s always value to just being a human resource in the newsroom that reporters feel comfortable talking to, and who prioritizes the same things that the reporters do.”

For the first four or five months of his fellowship, Aaron simply listened carefully to what journalists needed, rather than rushing to define a product.

“I don’t see my job as producing this tool. I see my job as producing reporting,” he said. “I just went to reporters and said, ‘I’m here to help you produce stories using AI.’”

That unstructured collaboration became the engine for the product he’s now building.

What he is paying attention to right now

Aaron is closely watching how human‑computer interaction is being reshaped by AI, especially the shift from code to natural‑language interfaces.

His Google Sheets add‑on for reporters is a concrete expression of that idea: Instead of writing code, reporters write prompts that operate over rows in a spreadsheet.

At the same time, he is focused on preserving agency and structure in reporting workflows, rather than handing everything to a black‑box agent.

“We don’t want a chatbot to just do everything for you. We want you to import your documents and maybe use a regular old Google Sheets function, and then the one little thing that AI needs to do, that a function can’t do, that’s where you apply it,” he said.

What he is most optimistic about

Despite the turbulence in the news business, Aaron is optimistic about the enduring value of human judgment in journalism.

“I’m optimistic that there are a lot of really smart, really determined people working on these problems,” he said. “I’m optimistic that the news industry is putting a lot of resources towards answering those questions.”

He believes there will undoubtedly be a market for automated news, such as hyperlocal weather reports. But he doesn’t see that eliminating the need for human reporters.

“There’s still a lot of demonstrated value in a real human telling you what’s important, a person who has dedicated their life to covering a subject, helping you understand these complicated systems that govern our world. That voice is still very important, and I see nothing yet that has obviated that.”

The Fast Four

A book, article, or newsletter you’d recommend?
The writing of Geoffrey Litt, especially his essay comparing AI to teleportation.

What’s one thing you’re learning that has nothing to do with AI?
New baking techniques.

What local journalism challenge deserves more attention?
Direct connection and human voice in an age of information overload and AI mediation.

One word that describes the next year of AI in news?
“Agency — the human kind. The more you can just use AI to do anything, the more important human agency becomes in making decisions about what you spend your time on, what you think is important, what you think should be communicated to a reader.”