If my earlier blog on the industry’s open-source moment is about why media needs to collaborate more, Schibsted offers a compelling example of how to build differently.

The company’s recently announced open-source text-to-video tool didn’t emerge in isolation. It’s the product of a much broader system — one designed to turn AI experimentation into scalable products. At the heart of this is a clear framework.

As outlined in a presentation at the recent INMA Agentic AI Master Class, Schibsted organises its AI efforts across three horizons:

• Horizon 1: efficiency and cost reduction.

• Horizon 2: evolving the product experience.

• Horizon 3: long-term strategic bets.

This creates focus. Not everything is treated equally, and not everything needs to succeed. Because if we wait for perfection, we may miss the opportunity entirely. Instead, Schibsted operates on a system of “bets.”

Small, fast-moving experiments run over three to six weeks with continuous evaluation based on ease and impact. The goal is simple: Maximise learning while minimising cost. This is how speed is operationalised.

But experimentation alone isn’t enough. What makes Schibsted particularly interesting is how they’ve built the infrastructure to support it.

Their AI Foundations layer acts as a shared platform across the organisation:

• LLM serving and endpoints.

• Model registries and feature stores.

• Workflow orchestration.

• Developer self-service tools.

This is critical. Instead of every team reinventing the wheel, they build on a common backbone — enabling reuse, consistency, and scale.

And then there’s the product layer.

Their “liquid content engine” is perhaps the most powerful concept. As I’ve written about before, liquid content is something every news organisation must understand and embrace. A single source-of-truth story is ingested, broken into atomic units (facts, quotes, visuals), and then reassembled into multiple formats — video, audio, newsletters, social.

This is not just automation. It’s a rethinking of content itself. Content becomes modular. Flexible. Reusable.

The text-to-video tool Schibsted is now open-sourcing is one part of this system. And this is where the open-source decision becomes even more interesting. By sharing part of this pipeline externally, Schibsted is effectively:

• Reducing duplication across the industry.

• Accelerating adoption of new formats.

• Positioning itself at the centre of a potential ecosystem.

It’s a classic open-source play, similar to what we’ve seen in tech.

But it also raises a bigger question. If one company can build this kind of infrastructure internally, what could the industry achieve together?

The real opportunity isn’t just better tools. It’s shared foundations combined with differentiated experiences on top.

Schibsted shows us what’s possible when AI is treated not as a feature but as a system. The next step is deciding how much of that system should be built alone — and how much should be built together. 

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

Banner photo: Adobe Stock By piter2121.

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