As AI evolves, it continues transforming newsrooms in new ways, and during this week’s Webinar — presented by the INMA Newsroom Innovation Initiative — members were given two revolutionary views of that transformation. 

Although news media companies have been experimenting with AI for the past few years, the approach has changed, said Florent Daudens, founder of Mizal AI and former head of AI at CBC/Radio‑Canada.

“AI is becoming a core infrastructure in the way we produce content and distribute it," he said. And that has changed how journalism is produced, distributed, and consumed.

The industry is moving far beyond ChatGPT‑4 as a tool. And instead is “building real AI systems.” He pointed to last year’s Pulitzer‑winning investigations, which didn’t use ChatGPT to take home the prize but instead leveraged AI for pipelines, embedding models, OCR workflows, and agentic architectures. used these tools.

“This tells you a lot about where we are,” he said. “We’re moving from tool to infrastructure.”

 The Web has evolved in three eras: Web sites, which gave way to mobile apps, and now is being handed off to AI agents — autonomous systems that act on behalf of users. These agents can fetch data, synthesise information, and generate outputs in any format: text, audio, or interactive conversation.

This shift, he warned, has three profound consequences for journalism.

“The first one is that the article will no longer be king because systems can fetch data that are being structured and produce an article or podcast or whatever based on this specific data points that they can fetch,” he explained.

Content will become conversational, multimodal, and personalised, because users will be able to interact with chatbots. But the most notable change is also one that should alarm publishers most: “The click will die, or at least be less important, because AI agents can open a browser, read hundreds of new stories and summarise them without ever clicking on your site.”

This rocks an industry whose distribution model has been built on clicks, this sounds an alarm.

“The click is how we built entire distribution models,” he said. “And it’s disappearing.”

Three pillars of AI as infrastructure

Daudens organised his talk around three pillars that define AI as infrastructure, explaining how publishers can use AI to leverage new opportunities.

1. Production

As a production tool, AI can surface stories that would be impossible to find manually. He cited Radio‑Canada’s investigative team, which used AI to sift through thousands of hours of audio from a cryptocurrency scam. “They basically were able to identify this person and it became the main character in the investigation,” he said.

He also highlighted how journalists are using AI to build tools that improve their own workflows, from Chrome extensions that check style‑guide compliance to interactive article formats coded by non‑technical reporters.

2. Systems

The second pillar is the shift from scattered prompts to reusable systems. Daudens emphasised the rise of “skills” — modular, plain‑language instructions that guide models through tasks like fact‑checking or chart formatting. These skills can be shared across teams, stitched together, and embedded into AI assistants that understand a journalist’s beat, style, and ongoing work.

The challenge, he said, is not technical but cultural. Newsrooms must learn to ship imperfect agents, gather feedback, and iterate — a mindset that clashes with journalism’s tradition of perfection before publication.

3. Distribution

The third pillar is distribution, where Daudens sees the most disruption. He pointed to an MCP connector built by the tech and business publication O’Reilly’s, which lets developers access its content directly inside their coding tools, and the Associated Press’s machine‑readable content service as early signs of a new ecosystem.

He also highlighted the Washington Post’s personalised AI podcast: “I think some people in the news are not happy with the product, and on the other hand, they produced more podcasts in three days than in their entire history because it was a podcast generated on the fly for users.”  

What publishers should do next

Daudens closed with three practical steps he encourages all publishers to take:

• Pick one investigation and run it through an AI document‑analysis tool to see what surfaces.
• Ask your team what your content would look like as an MCP server — even if you don’t build it yet.
• Install one journalism skill and test how it fits into your workflow.

Ultimately, he said, the newsrooms that build trust infrastructure for agents — and rethink distribution beyond the click — “… will be in the best position … for the revolution of how we access content.”

DRIVE reinvents local journalism with AI

As AI becomes the core infrastructure of modern newsrooms, regional publishers across Germany, Austria, and Switzerland are jointly developing AI capabilities together. Dr. Christoph Mayer and Dr. Ole Fehling of Heiberg Consulting shared the success of DRIVE, a unique collaboration of 30 regional publishers who are collectively accelerating digital transformation.

DRIVE is built on a simple but radical premise: regional publishers are stronger together than alone. Members share performance data transparently, benchmark against one another, exchange best practices, and co‑develop tools that would be too costly or complex for any single newsroom to build on its own. Not surprisingly, AI sits at the heart of this collaboration.

From convenience tools to transformation

Mayer began by reflecting on how quickly the conversation around AI has evolved. Just one year ago, when the INMA Subscription Summit discussed the most common AI use cases, “It was around headline optimisation,” he recalled.  

“Now the world has transformed so much,” he said. Yet in many newsrooms, AI is still used mainly for convenience — tools that make editors’ lives easier but don’t fundamentally change workflows or output.

DRIVE’s mission is to push beyond convenience and toward true transformation. Mayer and Fehling focused on two high‑impact areas:

1. Large‑scale automated content processing with AI.
2. AI‑assisted editorial work that multiplies the output and impact of human journalists.

To explain their approach, they introduced a pyramid that maps the layers of newsroom output, from routine community updates at the bottom to deeply reported investigations at the top.

Automating the “dark bread” layer of local news

At the base of the pyramid is what Mayer called “dark bread” journalism: police reports, school announcements, local events, and community notices. “No Pulitzer Prizes here, but a lot of sub-local content that is really important for the community and a layer of content where most newsrooms are not providing enough content,” he noted. These stories matter deeply to readers and are essential for regional publishers, but are also time‑consuming and often the first casualties of shrinking resources.

DRIVE saw the potential for this layer to be fully automated.

Their tool, Drive Local, processes incoming content — e-mails, attachments, PDFs, scanned documents, and council minutes — and turns them into publishable articles without human intervention. The system handles messy real‑world inputs, extracts relevant information, generates clean articles, and delivers them into the editorial CMS just like agency wires.

One example: a school concert announcement e-mailed to a newsroom. Drive Local transformed it into a finished article that appeared in the e‑paper edition without a single human edit.

The impact is dramatic. Publishers using Drive Local are producing over 1,000 AI‑generated articles per month, reducing manual effort by 80%. And freeing editors from low‑value tasks gives them more time to pursue deeper reporting.

The team is now expanding beyond e-mail to crawl regional portals, community council records, and other structured and semi‑structured sources — building a pipeline that can autonomously feed newsrooms with relevant, hyperlocal content.

Scaling the top layer

While the bottom layer can be automated, the top of the pyramid — comprising premium, deeply reported journalism — requires human judgment. AI still plays a role here, but that role is one of an assistant, not a replacement.

Fehling described a future where editors work alongside “multiple little AI helpers” that handle research, archive searches, data analysis, fact‑checking, and drafting. He showed an example of an AI agent navigating a computer, gathering local information, and drafting articles — illustrating how research can be streamlined and automated.

The real power, however, comes from DRIVE’s shared data, Fehling explained, introducing a tool called DriveMixer. The tool can easily navigate decades of article archives from the member publications, giving it access to editorial knowledge bases, performance data from millions of articles, user‑needs insights, and best practices across the 30 participaing publishers.

This data is exposed through an MCP server so that any DRIVE member can use it inside their preferred AI platform — whether that is Langdock, ChatGPT, or another enterprise tool.

Once the data is retrieved, editors can ask the system to complete any number of tasks, from drafting a chronology of events using archive articles to providing background context on breaking news to surfacing similar past incidents. It also checks names and facts against the archive. These capabilities allow journalists to produce more stories, richer stories, and stories they might never have found without AI‑driven research.

“We believe that for any kind of story and most of the tasks that editors do today, AI can really leverage what humans do,” Fehling said. “And it’s not by replacing humans, but it’s just helping them to do simple tasks like helping with research or drafting articles, fact-checking and so on.”

Banner photo: Adobe Stock Pungu x.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .