September 2026. It was a Tuesday, just before lunch, when an email landed in my inbox from a colleague with some thoughts about potentially valuable uses of AI in the team’s workflows. They asked for help. I asked one of my team members to follow up with a call, both to understand the proposals better and to ask whether they had tried anything with Copilot. ‘Copilot, do we have access to that?’ they had asked.
At first, I wasn’t sure I had heard correctly. Copilot? This is the AI tool with a capital A that was rolled out to all employees more than two years ago. The enterprise solution everyone is expected to learn through a series of training modules. Our natural starting point when superusers demonstrate best practice. That some journalists complain loudly about and others love.
Of course, this isn’t about any one colleague. I’m sure they use AI every day in all sorts of ways, even if they don’t happen to use that particular tool. But it is an illustrative starting point for reflecting on the state of AI adoption in media organisations. AI is on everyone’s lips. But how, when and where AI should be used at scale in editorial work – that is a nut we haven’t cracked as an industry.
Like my co-authors, I am convinced that AI technologies – used creatively and responsibly – can reinvent both the audience’s news experience and our internal ways of working. But before imagining possible futures, I want to begin with a self-critical reflection on my six years working with AI in journalism, drawing on leadership roles in editorial AI at Swedish Radio and BBC News. As usual, I will exaggerate slightly to emphasise my points. So please bear with me, and I hope you recognise some of these patterns from your own organisations.
Organisational archetypes and leadership blind spots
In all our feverish excitement about the dizzying opportunities of generative AI, I believe we forgot one thing: bringing the vast majority of our colleagues with us. In the past few years, I have come across four distinct types of colleagues when it comes to AI attitudes. A first group consists of those too tied up in daily production to have time to explore the new. A second are grumpily sceptical, as journalists always are. A third category are fierce opponents, although I believe that group is considerably smaller than some claim.
The fourth group is perhaps the most interesting: the pioneers, or ‘trailblazers’ as they are called in BBC terminology. Truly mind-boggling things were happening among this group almost from day one after the launch of ChatGPT in November 2022. For many, it was simply the next logical step after years of ‘hacking’ together their own digital solutions. But the best things cobbled up during 2023–24 were rarely brought into the light, at least not beyond their immediate circles. Sometimes this was due to a fear of getting caught using non-sanctioned tools or being accused of ‘cheating’. But fundamentally, there was neither a culture nor dedicated time for sharing those kinds of experiments.
I feel very much complicit in that. During my first few years in genAI, I focused a lot on theoretical possibilities and risks instead of spotting the concrete gems and giving them a bigger stage. Responsibility for practically exploring AI was left to the individual, whether they were at the cutting edge or largely indifferent.
There were attempts. I remember how Swedish Radio reporter Jesper Ericson showed how he had found an unexpected angle for an investigation into how schools in the Swedish county of Kronoberg compared with the national average. A textbook example of how AI can create editorial value. When he presented it at a management conference, people were fascinated. But back then we didn’t find an effective way to scale his methods across the wider reporter community.
Before I continue, I want to pause for a second and make one thing clear: in 2026, real progress is being made in broadening the use and integration of AI. Some media organisations are starting to crack it. Maturity is increasing. More people are getting involved. Real value is being delivered. But before we discuss that, let us dwell a little longer on the – and often my own – sins of the past. We will get to the forward-looking part, I promise.
Confessions of professional misconduct
This may sound shocking, even like gross professional misconduct, but in my early days in AI leadership, I very rarely took the time to talk to people further down the organisation about their needs, fears and dilemmas. Things were made worse by my move from an operational management job into a central strategy role, which meant I no longer had daily contact with frontline journalists.
So I rushed around different leadership forums and international conferences, knocked together slides and talked to like-minded people. I don’t think I have ever been as inspired and professionally happy as I was during those years. But did I spend enough time trying to understand my own colleagues? Hardly.
So what did we AI optimists miss during the first two or three years after the breakthrough of generative AI? To start, we enthusiastically jumped on the coolest innovation ideas. The shiny new things. ‘Now we’re going to launch new summary formats’ – which the audience hadn’t asked for. ‘We’ll let people ask our archives any question they like’ – but doesn’t Google do that better? ‘Look how easy it is to create illustrations’ – oops, that character seems to have six fingers on one hand.
Enthusiasm, improvisation and pioneering spirit rather than strategic focus. Quite understandable. We were sucked into the global super-narrative that the revolution had arrived and the race was on. Boards and management teams shouted: ‘Do something!’
The policy trap
Speaking of strategy – or rather the lack of it – many of us in the first phase of AI focused on guidelines. Media organisations spent much of 2023 and 2024 developing policies, all emphasising the importance of keeping a human editor in the loop. New committees and AI councils were established. Some of that principled editorial policy work was absolutely necessary at the time as a cornerstone of responsible AI work. Not least in public service media, where maintaining the audience’s trust is everything.
But it took a great deal of time and effort. The internal guidelines in particular started from what we should not do, rather than what value the technologies could add. Nor did they go hand in hand with a clear direction on what was most important to invest in. No wonder journalists were put off when the basic message was ‘Be careful!’ rather than ‘Have a go!’
Guidelines are not strategy. It took me a while to understand that. Nor did it sink in, as I rushed ahead, that strategy also has to be matched with real resources for implementation.
The far-sighted leaders
Some leaders saw the bigger picture earlier than others. When VG’s then-Publisher Gard Steiro, at the Nordic AI in Media Summit 2024, was openly paranoid about the industry’s limited appetite for change, many attendees were rightly shaken. After all, he was the one who was simultaneously initiating some of the most radical projects.
Swedish Radio (SR) Director-General Cilla Benkö was also early in forcing discussions about strategy and adoption into the executive meetings. ‘How do we involve all our newsrooms across Sweden?’ she asked with a piercing look. She and Programme Director Sofia Wadensjö Karén asked me to put together a first, simple slide with a forward-looking message. The recipe for success was its simplicity: just three points:
SR will use AI to…
…improve and develop our journalism
…create better user experiences
…improve internal productivity
… throwing spaghetti on the wall
We had a direction.
That didn’t stop me and my team from continuing to rely heavily on gut instinct and focusing on flashy things we could boast about at media conferences and on LinkedIn. ‘We’ve started working with few-shot prompting instead of model training’; ‘We’ve built a knowledge graph for our metadata extraction’; ‘We’re going to help train a Swedish language model – even one for Sámi’.
Of course, that’s not the whole story. Projects were in fact launched that could create significant value over time, such as automatic transcription of podcasts and an interactive news search. But somehow, we never quite got the execution right. Instead of rallying behind the strategic direction, we kept tinkering around the edges.
It was easier that way. Involving hundreds of journalists takes a lot of time and effort. You have to move at lightning speed in the AI era, we had been told. Blue-sky thinking was more fun than grey reality. Moreover, our colleagues in Product struggled to squeeze all our ingenious proposals into their packed roadmaps. Once again, we were haunted by the curse of the central innovation team: not that much of the spaghetti actually sticks to the wall.
The news factory kept the lights on
Meanwhile, our fantastic journalists kept toiling away in the news factory. And it was hardly business as usual. A dramatic world of wars and political convulsions created an unprecedented 24/7 news flow. An insatiable hunger for new formats across ever more platforms. Tighter finances meant chairs disappearing from newsrooms. And there was still a long way to go on the winding road from analogue to digital media organisations.
‘What, you haven’t started prompting and personalising your GPTs?’; ‘Surely you have taken part in Copilot Chat Prompt Workshops: Go Deeper?’ No wonder colleagues became even more tired when some AI maniac from headquarters appeared on screen with their insinuating questions.
Then there was the issue of middle managers. To put it bluntly: they were forgotten even more than colleagues on the newsroom floor. The first years’ attempts to increase AI engagement effectively bypassed the middle ranks, going straight from the executive team to individual employees
Learning to master new tools, navigating responsible AI principles, keeping up with external trends – all of that was largely left to each individual. At most, the middle manager might be expected to nag people into attending a few mandatory courses. But there was never any expectation for them to develop a plan for integrating AI into their team’s core processes.
Signs of bottom-up change
It is difficult to pinpoint when greater self-awareness began to set in. It was probably a gradual process, like a stalactite forming drop by drop. I had an early moment of illumination in spring 2024, when we asked Swedish Radio employees what they would use a perfect AI assistant for in their everyday work. The question was tucked away in a free-text field at the end of our introductory course on editorial AI.
We had hoped for a few suggestions. Instead, 1,224 (!) ideas came in. Many were considerably smarter than anything we could ever have come up with in our central AI team. The suggestions were clustered using AI. There were loads of practical ideas for journalistic work – research support, editorial feedback and audio processing.
But one other area was by far the biggest: administration. Colleagues at every level were stuck in the routine handling of information: reporting, email management and scheduling. It reinforced the impression that all employees desperately wanted to free up time. Had we focused too much on the editorial side?
The shared ideation process led to several tangible projects, such as a bot for HR questions. But we were still stuck in a mindset of running centralised projects. I had yet to realise that real empowerment might lie in radically decentralising the development of AI solutions.
Adoption challenges on a global scale
Perhaps it wasn’t until I joined a truly global media organisation in spring 2025 that I began to see the formidable challenges of scaling AI adoption. If it had been difficult to reach 26 local channels at Swedish Radio with information and activities, I had seen nothing yet. BBC News, with 4,200 journalists around the world, operates across so many time zones, formats, rhythms and production tools that engaging everyone is a Herculean undertaking.
At the same time, I was deeply impressed by the ambition of the BBC projects, often carried out with our R&D department. An in-house model for reversioning local democracy articles. A daily podcast covering 20 football clubs, where AI generates a script based on articles and a synthetic voice presents the content in the right local accent. A bespoke translation tool aimed at 43 language services around the world.
But switching on a new feature for thousands of busy journalists does not mean everyone starts using it. Launch and actual use are two completely different things. Even if we organise mandatory training and lunch-and-learn sessions, it doesn’t mean everyone takes part – at least not actively. To make a crude but affectionate generalisation: journalists are news avoiders when it comes to internal information.
The BBC had also, on several occasions, published critical research into how the tech giants’ AI assistants distort our journalism. It was a very valuable contribution to the debate about the lack of accuracy and attribution in the AI-mediated ecosystem. But if AI models from California are seen one day as striking at the heart of our mission, it is quite natural that reporters won’t rush to embrace Copilot, ChatGPT or Gemini the next.
Investing in newsroom change
My bosses Nathalie Malinarich and Anjali Kapoor began to see this: AI use was not increasing at the pace we wanted. At least not according to the way we were measuring it. We needed a proper strategy, linked to an action plan and greater capacity to implement it. Our growing project portfolio already spanned everything from video reversioning, style guides and newsletter support to AI voices, but now a new strand was added to the News AI team’s remit: adoption and scale.
In a short space of time, my hard-working colleagues set up an ambitious superuser programme. When I heard the participants’ comments in a workshop in London more pennies started dropping, one by one: ‘We don’t know what we’re allowed to do’; ‘AI is almost never discussed in the morning meeting’; ‘We want more practical examples of how AI is being used by different teams across the BBC’; ‘Copilot is getting better, but it’s hard to use for journalism.’
At the same time, I was overwhelmed by the engagement and precise ideas of the 40 participants. Once the magnifying glass was pointed their way, good examples started springing up like mushrooms. A colleague who had prompted their way to a subtitle-reformatting tool with an intuitive interface. A journalist who had scraped and transcribed hundreds of hours of YouTube videos to uncover stories and find people directly affected. An early version of an agentic chain that laid the foundations for an investigation with major impact.
Scale requires sustainable infrastructure
But, symptomatically, many of the most ambitious colleagues had built their tools using home-made solutions. Creative and impressive, but makeshift and not sustainable in the long term. The combined burden of barriers to accessing technology, legal issues, information security and governance made the process cumbersome, particularly because the innovators often had to try to navigate the internal bureaucracy on their own.
In a newsroom, having clear and easily navigable editorial systems is crucial. When I started publishing online in the early 2000s, you almost had to be an engineer to get an article onto the website. Things slowly improved as UX and data flows became more refined. Attempts were made to merge publishing systems as we moved towards a digital-first mindset.
But in the first genAI phase, we found ourselves back in a world of proliferating tools, with expensive licences for new third-party products. Without the appropriate experience, these immensely powerful general-purpose assistants are difficult to use for specific and repetitive tasks in a newsroom. Here I have been impressed by organisations such as Austria’s ORF, Denmark’s Ekstra Bladet and Finland’s Yle, which understood early on that their own AI toolkits were central to broad adoption. A bespoke interface with the company’s own branding gave journalists the seal of approval they needed to feel safe.
Perhaps this is my most important lesson for the next phase: we need to bring AI to journalists, rather than the other way around. Much like news audiences, I believe journalists are generally more interested in the user experience than in the sophistication of the underlying AI models. The technology needs to adapt to their needs and workflows, not force them to adapt to the technology.
The hard work starts now
So how do we equip our organisations for the wave of disruption that has only just begun? First, I hope my retrospective has provided some clues to the challenges we need to address as we move up the AI maturity ladder. Unfortunately, there is no magic wand. Swedish Radio’s Lead for Editorial AI, Danina Mahmutovic, has spoken about three elements that must go hand in hand: easy-to-use interfaces, hands-on coaching and continuous skills development.
But the day-to-day slog has to be paired with a visionary discussion about what journalism could become. That open conversation needs to take place – as it does in this anthology – across industry and disciplinary boundaries, and also within our own organisations.
For my own part, I believe that AI’s greatest opportunity in journalism is hyper-personalisation, both for news consumers and journalists. In terms of audience value, we have a real chance to offer content and services that are genuinely relevant to individual users. We can finally move away from mass media, which publishes the same content for everyone, towards a highly tailored offering based on interests, intentions, behavioural patterns and demographics. Is this the moment when public service media’s eternal dream of true universality can come true?
If we succeed, we can put real human needs at the centre in an entirely new way, with content coming to people rather than consumers having to buy physical products, turn knobs on appliances and navigate complicated apps. The media experiences of the future will evolve seamlessly with people throughout their lives. Access to vital, trustworthy information could increase dramatically.
It may sound utopian, but something similar could also emerge in editorial workflows. AI support could adapt organically to the individual journalist – based on their role, task, abilities and needs. It could become an ever-present personal coach, rather than yet another inflexible standardised system. AI could become a shortcut to the smooth digital workflows we have dreamed about for years but failed to achieve with traditional production tools. A bit like how some developing countries leapfrogged straight to mobile phones without taking the detour via expensive fixed-line telephony.
Glimpse of our possible future
Earlier this year, I caught a glimpse of the hyper-specialised editorial workflow when my colleague Annika Grosser showed me a prototype of a multi-agent system she had built that scraped, summarised and translated information from social media in a dictatorship. This provided material for a potential investigation. The workflow was carried out in a well-designed, bespoke tool – in practice, a vibe-coded CMS for that specific story.
I was dizzy with the possibilities. Imagine every journalist being able to build and use their own tools at every step of the journalistic process: from idea and research to production, publication and traffic optimisation. Instead of all digital development having to pass through procurement processes and battles over priorities, the person who actually has the idea and the right tech can start creating the solution themselves. The code monopoly would finally be broken.
But the flip side of the coin is darker. An AI leader at one of the world’s largest news organisations recently told me: ‘We’re now rolling out Claude to everyone and it’s incredibly exciting. But how we’re going to make sure everyone follows our journalistic guidelines when they build their personal agents, we haven’t got a clue.’
And it’s not all sunshine when it comes to the audience’s news consumption either. We compete, coexist and collaborate with the most powerful technology companies the world has ever seen. Is there any chance at all that people will be reached by the quality journalism we seek to provide? Will anyone recognise that a particular piece of content comes from us? Will hyper-personalisation on steroids create even more social fragmentation?
It is important to ask ourselves what it is we are trying to protect. No, it is not particular article, video or audio formats that are at the heart of journalism, as my fellow contributors Nikita Roy and Shuwei Fang often eloquently point out. Nor is it the media organisations themselves; more are likely to disappear in the coming years.
But good journalism is characterised by certain values: careful, methodical work with facts, a critical mindset, engaging storytelling, openness to different perspectives, an ambition to include underserved groups, and accountability when we get things wrong.
Journalism cannot be done entirely from behind a computer screen. Out there in the physical world are the talkative sources, eyewitnesses to dramatic events, and those affected – all absolutely essential to making stories come alive and hit you in the gut. If all we do is report ephemeral information, we’re toast. As warnings grow about deskilling – the gradual loss of professional skills – I’m less worried about us forgetting how to write than losing our ability to interact with people out in the world.
What needs to change to safeguard the core values
So what do we need to change organisationally to step things up? First, AI must become an integral part of every transformation process, rather than being treated as something separate and alien. That need becomes even clearer as we now introduce concepts such as ‘agentic AI’ without being able to explain what it means for the organisation, or even exactly what the term itself means.
It might be a good idea to start fewer office conversations with the diffuse and loaded term ‘artificial intelligence’ and instead...
…talk about enablement – giving people the chance to engage with new technologies – rather than about adoption, which casts people in a more passive role.
…focus more on local usefulness than on centrally developed, ready-made solutions.
…let go of our fixation on formats and artefacts and talk more about values and mission.
…make the strategic direction clearer rather than placing responsibility on the individual. Here, centralisation is actually needed.
…share more inspiring examples from peers who have created clever solutions.
…connect journalists with specialist expertise elsewhere in the organisation.
It’s the people, stupid!
Speaking of people. For 25 years, I have worked in different media organisations – print, radio, television and online media. At every point, some kind of crisis has loomed over us. First it was falling print revenues and listening numbers, then the threat from social media platforms, and now the rise of AI. Every organisation I have worked in has felt small and weak compared with the giants ‘over there’. All have gone through painful reorganisations.
But amid the turmoil are the people, our colleagues. If you are an inspired and hard-working bunch who want to do something together, you can move mountains. And if you don’t quite make it all the way, at least you’ll have tried, and hopefully had fun along the way.
So don’t wait. Connect with each other and get cracking. And please, feel free to learn from my professional sins whilst I get busy making new mistakes we can talk about next time.