TL;DR
AI Week in Milan is Europe's biggest AI event: around 25,000 people and 250 exhibitors, most of them selling the same handful of models with a logo on top. Three things were worth the trip, and all three matter if you run a business: AI that now writes and repairs its own code, the infrastructure hiding behind every chatbot, and a continent worried it is falling behind. Here is what each one means in practice.
AI Week is the biggest artificial intelligence event in Europe. On the 19th and 20th of May 2026 it drew around 25,000 people, 700 speakers, and 250 exhibitors across 17 stages. I went for one reason: to separate what is actually shipping from what is being sold, then bring the useful part back to the businesses I work with in Malta.
AI Week 2026, Rho Fiera, Milan
Most of the show floor is theatre
Robots and Cybertrucks pull the crowds. A humanoid robot walked the main aisle while people filmed it. Tesla parked its Optimus robot on a plinth next to a Cybertruck. It is a good show, and it is mostly a show.
A humanoid robot working the main aisle. Good theatre, not the point.The useful skill at an event like this, and on the next vendor call that lands in your inbox, is telling the substance from the wrapper. Of the 250 exhibitors, plenty were the same large language model everyone else uses, rebadged and put behind a monthly fee.
One question cut through it all weekend: what exact job does this do, and what happens when it gets it wrong? A stand showing AI that reads supplier invoices and posts them to your accounts could answer both. A booth selling an "AI growth engine" never got past the slogan. Same hall. One solves a problem you actually have. The other sells you the word "AI."
A Tesla Cybertruck on the show floor at AI Week
25,000+
People over two days
700+
Speakers
250+
Exhibitors
17
Stages
What the keynotes were really about
The big-stage talks were about scale. One walked through the data centres behind modern AI: the racks, cooling, and networking that every chatbot quietly runs on, billed as the building blocks for the fastest time to market.
A keynote on the data-center hardware behind modern AI
Here is the read that matters for a smaller business. You are not going to build that, and you do not need to. The model is a utility now, like electricity: owned by a few very large companies and rented by everyone else. That is freeing, not limiting. The expensive part is already paid for. Your advantage is not the model. It is your data, your workflow, and how well the AI is wired into them. Two companies can rent the exact same model and get completely different results. The difference is the wiring, not the engine.
The speakers backed up how serious this has become: Llion Jones, a co-author of the research that started this era of AI; Karen Hao, who wrote the definitive book on OpenAI; Lucilla Sioli, who runs the European AI Office. This is not a trend waiting to pass.
The real shift: code that writes and repairs itself
The talk I came away thinking about was "Agentic Self-Healing Code: the Death of Vibe Coding and Other Stories," by Jesus Garcia Hernandez, Global Head of AI at SDG Group.
Self-healing code is software that fixes its own problems. The AI writes the code, runs it, reads the errors when it breaks, and patches itself, in a loop, without a person pressing the button at each step. That is what "agentic" means: the AI acts, checks its own work, and corrects course on its own. It is a real jump from AI that just autocompletes a line for a developer.
His argument was that this kills "vibe coding": shipping whatever an AI generates without anyone who can read the code checking what is underneath. It feels like magic for about a month, then it hits a wall.
I review these codebases for a living, so I recognised every example. The pattern repeats: the demo works, the founder is proud, and the problems stay invisible until a real customer trips over one. An app that runs fine for fifty users falls over at five hundred. A booking form quietly takes double payments. A small change breaks two things nobody knew were connected, because no human ever understood the code. I have written before about what happens when a vibe-coded app stops scaling and the security holes these tools leave behind.
Here is the part worth keeping. Better AI does not shrink that risk. It grows it. When a machine writes ten times more code, ten times faster, the gap between "it runs" and "it is safe to run a business on" gets wider, not narrower. Self-healing tools close some of that gap on their own. They do not replace the judgment about what to build, what to test, and what to never ship without review. The cheap part got cheaper. The judgment got more valuable.
Europe's real problem is not the technology. It's adoption.
One theme came up on stage after stage. Europe is behind on AI, and the worry is less about who builds the biggest model than about who actually puts it to work.
A Brembo Solutions keynote on the formula for scaling AI adoption
The comparison was always the same. The United States has the models and the money. China has the scale and the speed. Europe has talent, rules, and a habit of waiting. The fear is a continent that rents its AI from abroad and never builds the muscle to use it well.
At your scale the lesson is simpler. A firm in Texas or Shenzhen is already automating its boring work. A firm of the same size in Europe is often still deciding whether to begin. That gap is not about budget or genius. It is about starting. The technology is on the table, priced for anyone, and the advantage goes to whoever picks it up first. Malta included.
What this means for a business in Malta
Strip away the robots and the keynotes, and three things are true today.
AI is good enough to do real work in your business. The invoice that reconciles itself, the inbox that triages itself, the quote that drafts overnight. This is not a someday promise. It runs in real companies right now, and the ones using it well are getting hours back every week.
The shortcut is the trap. A cheap prototype that looks finished and falls over the first time it meets a real customer costs more to repair than it would have cost to build properly. Spend the saving on foundations, not on launching a day sooner.
Waiting is the real risk. Not a competitor's new model. Your own delay. Every week the boring work stays manual is a week someone else used to pull ahead.
Where to start
You do not need a 25,000-person event to act on any of this. Pick something small and let AI take it off your plate.
- Pick the task you hate most. The repetitive one that eats an hour a day. That is your first candidate.
- Start with one workflow, not ten. One inbox, one report, one form. Prove it, then expand.
- Keep a person in the loop. Let AI draft, let a human approve. That one rule prevents most of the disasters.
- Build it to last, not to demo. If it touches real customers or real money, treat it like real software from day one.
The companies that win the next few years will not be the ones with the flashiest demo in Milan. They will be the ones that took one real problem and solved it properly.
Thinking about putting AI to work in your business?
I build custom software with AI built in, designed around how your business actually runs. Not a throwaway prototype, something you can run on for years. Tell me what is slowing you down and I will tell you straight whether it is worth building.
