How newsrooms move from AI pilots to infrastructure

We’re starting the fourth year of generative AI experimentation. But still, according to Nic Newman's Reuters Institute Journalism and Technology Trends and Predictions 2026, only 13% of news executives describe their AI initiatives as "transformational." Another 42% call them "limited." Two-thirds report no job reductions from AI efficiencies. The promised savings haven't materialized (yet).

"In 2023-24, many news producers automated individual newsroom tasks," writes David Caswell, a British expert in applied AI for news products and workflows, in a January 2026 forecast for the Reuters Institute: ”By 2025, the limits of 'task automation' have become apparent. Savings of time and money are underwhelming, and task-focused AI seemed like a strategic dead-end."

Caswell argues that the next phase requires a fundamental shift - from automating individual tasks to deploying agentic AI that handles end-to-end workflow automation. The difference matters. Task automation speeds up discrete and individual steps. Agentic systems manage entire processes, making decisions and routing work without constant human direction.

But the cautionary tales from the task-automation era keep accumulating. Sports Illustrated published articles under fake AI-generated author names. CNET's AI-written pieces contained factual errors. Gannett faced social media backlash over automated content. The L.A. Times pulled its "Insights" tool after one day. In December 2025, the Washington Post weathered an internal revolt over AI podcasts riddled with errors.

Public trust reflects the skepticism. A Pew Research Center survey from spring 2025 found roughly 50% of US adults expect AI to negatively impact news over the next 20 years. Only 10% expect positive effects.

Against this backdrop, two distinct paths are emerging for newsrooms ready to move beyond pilot mode.

The New York Times represents one model: invest in a small, focused team that builds custom tools while changing how journalists think about AI.

Zach Seward, Editorial Director of AI Initiatives at The New York Times, estimates he has engaged with 93% of the newsroom over two years, starting by clarifying AI definitions and establishing generative AI guidelines before hands-on trials.

Reporters now routinely use AI for investigative work, such as organizing "massive messy data sets," including Trump administration officials' public statements analyzed with the Washington bureau.

His team shifted from isolated projects to tools like "Cheat Sheet" (for spreadsheet summaries of large document/image/transcript sets) and the "Manosphere" daily email (summarizing right-leaning male-targeted podcasts).

"People come in with a lot of existing thoughts and beliefs, which means it's all the more important that we're there in person," Seward told me in October 2025. "Some people are super excited, ready to go. Others are the exact opposite, understandably horrified by some of the things they see other publishers doing with AI.” The goal isn't converting skeptics into enthusiasts. It's finding specific, comfortable use cases.

To avoid pilot purgatory, Seward uses a method borrowed from Basecamp's Shape Up methodology: declaring an "appetite" rather than a scope. "Someone says, 'I want to explore small models for summarization.' I might say, 'That's interesting. I've got two days’ appetite for that.' That means go ahead and spend two days working on that, and at the end of two days, we're going to stop basically no matter what."

Scale thinking shapes every project. "You need to transcribe these videos, get them analyzed, and publish your article. We're going to focus on that. But in the back of our mind, it's like, 'It's not the last time we're going to be asked to transcribe a bunch of videos. So let's do it in a way that makes it easier the next time.'"

For smaller newsrooms, Seward suggests the barrier is lower than assumed: "You could get a really long way with one person dedicated, ideally full-time, but depends on the size of the newsroom. Maybe it's the most enthusiastic person in the newsroom who also has five other jobs."

News Corp has also run several successful AI experiments. Its flagship newspaper,  The Wall Street Journal, developed its own tax bot Lars. But on the systemic level of publishing it now went with a vendor. On January 15, 2026, the company announced a partnership with Symbolic.ai to deploy a unified AI publishing platform across its properties, starting with Dow Jones Newswires. TechCrunch reported the deal positions Symbolic as a potential industry standard.

Symbolic.ai was founded by former eBay CEO Devin Wenig who previously ran Thomson Reuters Markets, and Jon Stokes, co-founder of Ars Technica. The platform offers a workspace for transcription, document extraction, newsletters, fact-checking, headlines, and SEO optimization. It routes tasks to multiple AI models rather than depending on a single provider.

According to Symbolic’s press release, Dow Jones Newswires’ early use of Symbolic yielded productivity gains of as much as 90% for complex research tasks. 

The platform approach promises speed and standardization. Instead of building custom tools, newsrooms plug into existing infrastructure. For news organizations without engineering resources, third party platforms that orchestrate multiple AI capabilities into coherent workflows rather than offering isolated tools may be the only viable path beyond scattered AI experimentation. 

Upasna Gautam, senior product manager at CNN, offered a framework on the Newsroom Robots podcast in December 2024. Before implementing any AI tool, she asks three questions: What's your biggest goal? What's your most significant pain point? How do you measure success? "I never skimp on context and justification," Gautam said. These fundamental questions haven’t changed, but they have become more consequential when more than just the success or failure of an AI experiment in a sandbox is at stake.

The build-or-buy decision depends on honest answers. Organizations with unique editorial workflows, proprietary archives, or competitive advantages worth protecting may benefit from custom development. Those seeking efficiency gains across standard processes may find platforms deliver faster results.

Francesco Marconi, writing in Nieman Lab's 2026 predictions, framed the stakes: "2026 is the year newsrooms begin their first-principles rebuild for the AI era. It is the year we stop treating journalism as a content factory and start rebuilding it as a knowledge and community institution."

The experiments generated lessons. Now comes the harder work of turning those lessons into infrastructure, whether built from scratch or bought off the shelf.