Nicolai Vittrup

← All thoughts

Thoughts, 10 November 2026

Web Summit day 1: Everyone has AI. Almost nobody makes it work

Nicolai Vittrup looking out over Lisbon from a viewpoint

The first time I went to Web Summit was in 2017. Almost ten years ago. Back then, Stephen Hawking opened the conference by video and warned that AI could be the best or the worst thing ever to happen to us. Back then, AI was something you warned people about.

This year, AI isn’t a topic at the conference. It is the conference.

For those who don’t know it: Web Summit is Europe’s biggest tech conference, and it’s held in Lisbon. I’d decided that today was only for learning. No meetings, no networking, just talks. I ended up at nine in a row. The rain blew in sideways, it was windy, and there was a queue outside every single hall. When more than 70,000 people are gathered in one place, that’s just how it is.

Queue in a tent outside a hall at Web Summit in Lisbon
The queue for the masterclass on AEO. There were queues for most things all day.

The gap to Denmark

Web Summit is about what’s coming. That’s why I come down here. You see how big the gap back home is.

In Denmark, we talk a lot about regulation. Down here, they talk about everything that’s already changing. The agents are coming. Marketing is changing. The media are changing. Even Formula 1 is changing.

Atlassian Williams F1 Team were on stage talking about how they’ve rebuilt all their technology. The team principal, James Vowles, also described how he found out that the 2026 car was in trouble. It wasn’t the numbers that told him. It was the silence:

“All of a sudden, the information stopped coming. And it was the first sign that we were really in a problem.”

Nobody wants to go to the boss with bad news, he said. And they had nothing that showed the problems until it was too late.

That’s one I’m taking home as a leader. When you don’t hear anything from your team, it isn’t because everything’s running smoothly. It’s because it no longer feels safe to say when something isn’t working. It’s not the technology that’s failing. It’s the culture and the way people work together.

And that was exactly what the rest of the day was about. How we work with clients, how our own people work, and how you build a company in a new way. That’s what I call an AI-first agency, and I got more to think about than I’d expected.

Nicolai Vittrup sitting against a pink wall in Lisbon
Lisbon, the day before it all kicked off.

When nothing joins up

The talk that gave me the most to take home was by Dan Gardner of the American agency Code and Theory. They have more than 2,500 people, half creatives and half engineers. He started by saying he usually opens with a joke. He didn’t need to this year, because the whole situation is a joke.

Generative AI has been around for four years, and all his clients ask the same question: why doesn’t all this AI add up to more than the sum of its parts? His answer is what he calls “the orchestration gap”. In short: people, tools and data don’t work together. Every department does its own thing.

Together with Wall Street Journal Intelligence, Code and Theory surveyed more than 800 top executives at companies with revenue of over half a billion dollars. 94 per cent say they’d grow more if people and tools worked better together. Hardly any of them have done anything about it.

Dan Gardner of Code and Theory on stage at Web Summit
83 per cent of top executives think the customer experience is good. 75 per cent of their own people say the data doesn’t join up.

A few of the numbers:

  • 183 tools on average. 13 per cent have more than 500.
  • 25 per cent have nobody responsible for making it all join up. And where there is someone, the CMO, the CTO and the CEO all think it’s them.
  • 83 per cent of top executives think the customer experience is good. 75 per cent of their own people say the data doesn’t join up.
  • 70 per cent have no set plan for getting found by AI agents.

These are big corporations. A Danish company with 20 employees doesn’t have 183 tools. The problem is the same: every new tool solves one thing and makes the whole thing a bit harder to hold together.

His conclusion was simple:

“We don’t need more tools doing individual things. That is not solving the problem. That’s almost making it worse.”

And then he said one sentence twice, so everyone got it:

“Technology is not the limitation.”

25 years ago, technology was what held us back. Today he can’t think of anything technology can’t do. What holds us back is the way we work.

“AI doesn’t care about your org chart.”

And then came the line I’m taking home above all others:

“AI’s best value is compression, not automation.”

It’s not about automating. It’s about making the path shorter. From idea to customer. Between departments. From a good idea in one corner of the company to everyone else being able to use it. If all you’re after is saving time, you end up at the bottom of your industry, he said.

He was also hard on all the talk about upskilling people:

“Upskilling is a joke.”

People can use ChatGPT just fine. What’s missing is new ways of working, and getting rid of the silos. I very much agree with that. He also took on “human in the loop”, which everyone says and nobody can explain. Is the human just the one who now has to approve a thousand things instead of ten? Then it gets slower, not faster.

Slide reading: AI can't act on context it can't access
AI can’t use knowledge it doesn’t have access to.

What it means for us

At Amplify, search is still a big part of what we do, both on Google and in AI. And Gardner’s talk is the one I took the most notes from.

We’ve built our own AI agent. The best thing about it isn’t the individual tasks it handles. It’s that all the knowledge is gathered in one place. Normally, what you know about a client is scattered across emails, Slack, meeting notes, ad accounts and spreadsheets. Our agent pulls it all into one picture of the client that we and the agent both work from. The rules the team has approved live in the same place, with the reasoning next to them. Every Wednesday, for example, the agent checks whether people are getting tired of seeing the same ads on Meta, and flags it.

That’s exactly what Gardner is talking about. Not one more tool. Something that ties them together. We already do that, and we’ll be doing even more of it.

Can small businesses keep up when agents do the shopping?

Matthew Prince is CEO of Cloudflare, which is behind a large part of the internet’s infrastructure. He made the most uncomfortable point of the day. The internet has lived off ads for years, and:

“Agents don’t click on ads.”

That wasn’t even what worried him most. It was the small businesses. Maybe you use the dry cleaner on the corner because it’s right there. Or the butcher, because you know him.

“Your agent doesn’t care about either of those things.”

The agent chooses based on what it can find online, and that favours the big players. In his worst-case scenario, the internet as we know it stops existing if nobody finds a new way to pay for content.

Matthew Prince of Cloudflare on Centre Stage at Web Summit
Matthew Prince on Centre Stage. Agents don’t click on ads.

He’s right about how it works. It’s easier to be big when more people are writing about you. When people search in ChatGPT or Google’s AI, the big players get mentioned far more often.

I still think everyone has a chance if they do it right. It just means the small ones have to think much harder about what they publish, so they get found both on Google and by the agents.

Michael Komasinski, CEO of the advertising platform Criteo, showed why it’s worth fighting for. According to their figures, 80 per cent of the people who come from ChatGPT are new customers for the brand. And they end up buying 1.2 to 1.5 times as often as people who come from an ordinary search. He had the best one-liner of the day:

“Visibility earns you consideration; relevance is what shapes choice.”

Visibility gets you on the list. Relevance decides whether you get chosen.

Criteo slide on AI as the new front door to e-commerce
Criteo: AI is becoming the new front door to e-commerce.

Content is king. Again

If I were sitting across from a small Danish company tomorrow, I’d tell them one thing: content. Original content, new content, both text and video. And get others to write about it.

“Content is king” is something people have said for 20 years. The difference is that now there’s an AI reading along. It loves new knowledge, and it goes much deeper into the material than you and I do.

Isobel Sita-Lumsden from TikTok said the same thing from the other side. People no longer just search for something specific. They stumble across it. And then the content has to earn the attention:

“Stop thinking about interrupting attention and start thinking about how to earn it.”

Helen Martin from Profound, which measures what AI services say about brands, explained it better than I can. She started with the question everyone asks: do we show up in ChatGPT?

“Is my brand the answer? But that’s actually only half of the question.”

The other half is what the answer says about you. They’ve gone through 50,000 answers from the AIs.

Helen Martin of Profound presenting a slide showing that AI answers are long
AI answers aren’t short. They’re long and full of extra recommendations.
  • On average, an answer is around 3,000 characters. That’s 11 tweets about your brand, even when someone just asked about a price.
  • 40 per cent of answers include something nobody asked about. Comparisons, explanations and suggestions for what you should buy instead.
  • The American airline Southwest was recommended for groups. And then you were told to arrive early, because there’s no allocated seating. Southwest dropped that rule more than a year ago.

So you can show up and still have the wrong picture painted of you.

“You have to be the API between ChatGPT and your brand.”

The AI knows everything that’s on the web. It doesn’t know what it feels like to fly. Or how your shoes feel at mile 18 of the London Marathon. That’s what you have to give it.

Profound chart showing that half of the most-cited content is less than 13 weeks old
Half of the most-cited content is less than 13 weeks old.

And then came the number that should make every marketing director take a look at their own blog: half of the content that gets cited most is less than 13 weeks old. A single good blog post used to sit at number one on Google for six months. That’s no longer how it works.

The best example was RunRepeat. A site that cuts running shoes open and tests them in a lab. In retail, they’re often the source AI cites most of all. Ahead of Nike, Adidas and Puma. Helen said herself that they probably don’t do it to get found by AI. They just love running shoes.

Profound slide comparing vague and specific content
Vague content gives the AI nothing to work with. Specific content is something it can compare and cite.

And a couple of good articles aren’t enough:

“To be the answer in AI, you have to be the answer everywhere.”

Reviews, social media, press releases. Everything others write about you counts.

That fits with what I wrote in GEO and AEO are about content. It’s the same craft as SEO, just turned up a notch: something nobody else has, spelled out in concrete detail and kept up to date.

What I’m not buying

Not everything you hear down here holds up once you bring it home to Denmark.

“The experimentation phase is over.” Gardner said that anyone standing on a stage telling people to experiment more is wrong. That makes sense for an American corporation with 183 tools. Most of the Danish companies we meet aren’t done experimenting. Many haven’t even started. For them, the advice is still: get started.

Every talk ended with a product. Code and Theory finished with their own platform. Profound with their AI marketer. Criteo with theirs. That’s completely fair. It’s how conferences work. It’s just worth keeping in mind when you read the numbers.

Tracking your ranking in ChatGPT. Profound sells tools that show what AI says about you. That’s good for catching mistakes like Southwest’s. And Jens from our team puts it pretty precisely: you can’t measure AI visibility on individual questions. The answer changes with the platform, with where you are and with how you ask. That’s why we’d rather look at how many people actually come in from the AI assistants, and how many of them become customers.

AI tools also get most of their knowledge from perfectly ordinary searches. So strong SEO is still the foundation. And what others write about you counts. If you’re not in the guides and comparisons when someone searches for “best” in your industry, those are the sources the AI listens to. It cross-checks sources. It doesn’t just rely on your own website.

What I’m taking home

Nicolai Vittrup in a street in Lisbon
The notes need to sink in before they turn into anything.

We’re not going to change everything when I’m back home. We’ve already come a long way, and most of what I heard today fits with the way we think.

We just need to do more of it:

  1. Bring more together in one place. Knowledge, tools and people need to join up, instead of us adding more tools side by side.
  2. More original and specific content. Things only we and our clients know, and that get updated before they’re three months old.

Gardner finished with a line I completely agree with:

“When attention changes and behaviour changes at the same time, disruption happens.”

It happened with the internet, with social media and with mobile. It’s happening again now.

Day 2 was cancelled because of the storm. So this turned out to be my Web Summit.

Centre Stage at Web Summit in Lisbon, seen from the auditorium
Centre Stage, late in the afternoon.

More thoughts