How the German Press Agency is reinventing news distribution for the agentic age
Founded in 1949, dpa (Deutsche Presse-Agentur) is a joint venture of about 170 German media companies with the mission to provide reliable information for media outlets, companies and organisations to use in their journalistic work.
dpa’s work has historically fallen into two key areas, said Yannick Franke, the press agency’s AI Team Lead, speaking at WAN-IFRA’s Frankfurt AI Forum.
First, the agency provides its clients with information about global news events through its wire service. Client companies can publish this content directly or build on it with their own additional reporting.
Second, the agency’s customers have access to a news hub platform, where they can seek information about a specific topic based on dpa’s coverage over the last few years.
However, Franke said this framework, which has guided the agency’s operations for its 77 years of existence, is now being upended by AI.
“The whole industry is changing,” he said, especially because “a lot of knowledge work and information work is now done not directly by editors and journalists, but by intermediates – by AI systems.”
If we are moving towards a world where AI agents do the bulk of information-seeking, Franke said this raises existential questions for dpa: “How can we fulfil our mission to inform media outlets and give them trusted information in a world that works so differently? And how can we support our customers, our owners, in their transformation processes?”
The agency’s answer to this starts with a clear “core premise,” Franke said: “Wherever the AI work of our customers is happening, we want to give them easy access to trusted information.”
To that end, dpa is soon launching a platform it calls dpa-iq, a “trusted information layer for agentic systems.” (It is currently in private preview.)
Serving agents with trusted information
Although dpa-iq is being created with agents in mind, it aims to replicate a use case that is familiar from the agency’s news hub, which journalists often refer to when seeking reliable information about a specific topic.
dpa-iq is designed to be a similar destination but for agents, Franke said: “an API product where agents or AI systems can get trusted information about specific topics.”
When an AI agent is tasked with finding specific information, it can visit dpa-iq, which will in turn “look for relevant information and provide this information to the agent, so that the agent can fulfil its task, do its value proposition,” he said.
A journalist could, for example, instruct an agent to seek specific pieces of information or other material, such as recent news relating to the situation in Iran, or an image of a politician in a defined location, or appropriate B-roll video content.
“Whatever information or items are needed, agents can come to dpa-iq to get relevant information,” Franke said.
Initially, the platform will include information exclusively from dpa, from articles and videos to images, databases and audio files. But the agency is in discussions with a number of partners about integrating them as additional sources in the system.
Sports data is one potential area for this: the users of dpa’s news hub often seek information about sports, but “the problem with sports questions is, most of the time those are data questions. And we don’t have a dpa article for every data question,” Franke said.
Additionally, dpa will be including information from a data provider that collects data from various German government bodies, structures them, and makes them available at different geographical levels.
Multi-source retrieval service
A key goal for dpa-iq is to enable product teams – both internal and external – to use it as a foundation for building additional products.
To that end, as an API management platform dpa-iq will include granular controls for multi-user situations, allowing access rights and rate limits to be defined for each individual user.
Moreover, because the tool is built as a “multi-source retrieval system” as opposed to “one big pillar of data,” Franke said it makes for a flexible, modular foundation on which new data sources can be integrated in the future.
The platform’s structure will also include a “multi-source retrieval endpoint,” where customers can ask questions and have the tool search various sources to find the relevant information.
Finally, dpa-iq will also include a “generation endpoint” that generates answers to questions. However, Franke was clear that their goal is not to build a chatbot: “This is not the core value proposition. This is just for commodity, to make it easier to deploy applications on top of that.”
Given the speed with which the AI space is evolving, the underlying infrastructure that powers dpa-iq could look very different even in the near future. That’s why designing a modular structure was so important, Franke said, as “we can just plug and play different vendors, different technologies, different services.”
Going back to the initial “core premise“ of bringing dpa’s content to wherever its clients are doing AI work, the team is further enabling this with upcoming integrations for various relevant platforms, such as the AI integration platform Langdock and OpenAI.
“Same goes for process automation tools,“ such as Zapier, n8n and Make, Franke said. The team has already built a demo for a solution using these tools, where at 6:00 am each morning the workflow scans the dpa-iq archive for specific information and presents the materials in a publication-ready newsletter.
