What's the most important question right now?

Speaking from a European public-service media perspective, I don’t think there is a single most important question right now — there are several, and they are closely connected.

We are living in a world shaped by large language models where information itself is no longer scarce. Search engines are turning into answer machines, and simply repeating what news agencies report — regardless of format — no longer creates real differentiation. This forces journalism to rethink its core value proposition.

This raises a set of strategic questions: how journalism can move from distributing information to building meaningful relationships with its communities and audiences; how trust, credibility, and relevance can be created through editorial choices rather than scale; and how audiences are treated as participants rather than passive consumers.

From a European perspective, this also means embracing plurality as a strength. Different societies interpret the same events through different historical, cultural, and political lenses. Journalism should not flatten these perspectives, but actively surface and compare them. AI can support this work — not by replacing editorial judgment, but by helping journalists contextualize, contrast, and explain complexity.

Another connected question is how newsrooms use data and modern tools to find stories that matter. In environments with accessible public data, AI-enabled analysis can (for example) uncover patterns around budgets, infrastructure, regulation, accountability etc. that would otherwise remain invisible. But this requires new skills, a different mindset, and a willingness to invest in investigative capacity rather than automation for its own sake.

All of this ultimately leads to a broader strategic challenge: reclaiming the digital public space. Today, much of public discourse happens on platforms governed by opaque algorithms and external interests. Journalism should see the public sphere as an editorial responsibility — creating spaces for debate and exchange where journalistic values, not platform logic, set the rules.

What will we be shaking our heads about a year from now?

How fast content outputs have become fluid, multimodal, and hyper-personalized — “rendered” and distributed on the fly for each individual user.

The speed itself will be striking, but also the direction it initially took. Driven by existing monetization and engagement logics, hyper-personalization will mostly be used to optimize for comfort: adapting tone, depth, perspective, and format so precisely that audiences rarely encounter friction, contradiction, or surprise. That will make already familiar dynamics — echo chambers and reinforced worldviews — even more powerful.

At the same time, this is exactly where the missed opportunity lies. The same technologies could allow publishers to reach people who never felt addressed by a “typical” media voice before. A single story could be told in different ways — simpler or more complex, short or long, visual or textual — without abandoning editorial identity or responsibility.

The decisive question won’t be whether content is personalized, but who controls that personalization — and whether it remains tied to a clear editorial identity. With humans in the loop and clear editorial guardrails, hyper-personalization doesn’t have to mean fragmentation; it can become a tool for inclusion and reach.

What we may end up shaking our heads about is not that this happened, but that it was so predictable — and that it still took us so long to develop a strategy, clear editorial principles, and the technological and organizational foundations needed to act on them before being overtaken by it.

What's a good website?

A website I genuinely love is Models All the Way Down from the Knowing Machines project. It explains how generative AI works at a very fundamental level — not with jargon, but with simple language, strong visuals, animations, and concrete examples. You do have to scroll quite a bit, but every step makes the next one clearer. What I find so impressive is how it turns something most people vaguely “know” — that image generators are trained on huge datasets — into real understanding.