Why It’s Easy to Launch Speedboats and Hard to Bring Them Home

Hi, I’m media innovation journalist Ulrike Langer and you’re reading News Machines. If you get value out of my newsletter week after week, please consider upgrading to the premium version. Premium subscribers get valuable bonus material at least once a month - including in this issue.

Every newsroom experimenting with AI runs into the same wall, and it isn’t a technical one. The experiments work. A small team builds something clever, it does what it promised, everyone is impressed – and then it stays where it was built, never reaching the core newsroom where most of the journalism actually happens. The hard problem in newsroom AI right now is not invention. It is transfer: getting what a fast, isolated team learns into a large, slow organization that wasn’t built to absorb it.

VG, the Norwegian tabloid and one of Europe’s most openly pro-AI newsrooms, has gone further than most at confronting this. Speaking at the Nordic AI in Media Summit in Copenhagen, its editor-in-chief and CEO Gard Steiro went as far as making his own automated app’s failures public. That candor turns out to be the most useful thing in VG’s whole AI story – but to see why, you have to start with the structure of the problem.

Steiro builds his case on an old observation from technologist Scott Brinker, sometimes called Martec’s Law: Technology changes exponentially, organizations change logarithmically, and the space between the two curves is unused potential. Training staff narrows the gap a little. It does not close it, or narrow it fast enough. Asked whether a large legacy organization like VG can retrain its way into moving at the speed AI now allows and requires, Steiro’s own answer is blunt: No.

That leaves a structural problem. You can’t keep the old article-by-article production line – it can’t match the scale and speed AI enables. But you also can’t simply retrain thousands of people into a new way of working before the opportunity passes. How does a tanker change course?

VG’s answer is to stop trying to turn the whole ship at once. Instead, it launches what Steiro calls speedboats: small, fast teams that run ahead of the main vessel, take sharp turns, and experiment freely without putting the core brand – and the audience trust attached to it – at risk.

The metaphor is appealing, and it is not unique to VG. At the same event, Politiken’s editor-in-chief Amalie Kestler described her newsroom’s options in nearly identical terms – heavy tanks versus light, fast drones. The model has a long history outside journalism, where it carries a well-documented weakness: The fast team produces something genuinely new, and the parent organization never manages to adopt it. The speedboat sails. The cargo rarely makes it back aboard.

The real test is not whether VG’s speedboats move – they obviously do – but whether anything they learn ever crosses into the tanker.

VG’s flagship speedboat is VGX, an app stripped of the things print-era news takes for granted: no fixed front page, no standard article format, and – the detail Steiro emphasizes – no content management system. It runs around the clock with no editors on shift. Underneath it is a clustering engine, built to be shared across VG’s parent company Schibsted, that scans incoming articles and video, groups them by meaning, and decides whether each new piece starts a fresh story or joins an existing one. A set of specialized agents handles the writing; an experimental layer lets an editor adjust the product by instructing an agent in plain language instead of editing a CMS.

Steiro is unsentimental about all of it. The user experience, he says, is not great yet. Asked whether VGX will survive as a product, he says it probably won’t. That sounds like a strange thing for a CEO to say about his own flagship, until you remember what the speedboat is for. It was never meant to become a hit app. It was meant to generate something the tanker could use – and on Steiro’s own account, exactly one thing has crossed: the clustering engine, now running on VG’s main product to give returning readers a personalized summary of what changed since their last visit. One transfer, from the one component deliberately designed to be reused.

VG’s other speedboat, VG Lab, chases business models rather than journalism. Its screening tool scores any idea – can VG build it, is there a market, what would it cost – and scans the world for concepts worth copying. A two-person team plus a set of agents can stand up a working prototype in a day or two. Steiro says the lab built the most-downloaded app in Norway last autumn and now produces millions of kroner in revenue that help pay for journalism.

The speed is real and the revenue is real. But measured against the transfer problem, VG Lab is doing something different from what the speedboat metaphor promises. It isn’t feeding anything back into the newsroom’s journalism – it’s a profitable startup running alongside the parent, an in-house revenue source rather than a course-correction for the tanker. A useful thing to own. Just not evidence that the model solves the problem it was built to solve. Across both speedboats, one reusable tool has made the crossing. Nothing has yet changed how the core newsroom works.

So how does an innovation get from a speedboat into the main operation? Pressed on this, Steiro gave the honest answer: It depends on a particular person spotting the innovation and carrying it into the main products by hand.

That is the whole mechanism. The step the entire strategy rests on – the crossing from fast experiment to core operation – is not a process, a pipeline, or an institution. It is one capable individual noticing something and walking it over. That is the whole mechanism. The step the entire strategy rests on — the crossing from fast experiment to core operation — is not a process, a pipeline, or an institution. It is one capable individual noticing something and carrying it over, by hand, every time. There is no permanent mechanism. There is a person.

The mechanism is fragile. But Steiro is refreshingly candid about weaknesses and errors. He put VGX's errors in front of a room of fellow newsroom leaders: a minor Swedish celebrity elevated to breaking news, the system’s own internal reasoning leaking into a published summary, a story rendered in an accidental Norwegian-Swedish hybrid language, the Swedish king mistakenly made head of state of Norway. 

Because VG showed the failures, the whole attempt is available to learn from, not just the parts that worked. The value isn't a solved problem; it's the honesty of showing it unsolved. They’ve done the genuinely hard things: launched real speedboats, let them fail in the open, and refused to pretend the flagship is a triumph. 

What they haven't built – what arguably no one has – is a system that moves innovations from experiment to core operation automatically. Launching the speedboat is the easy part. The crossing is what decides whether any of it scales.

Below the break: