What I saw at WMF 2026, why “AI wrappers” are the symptom and not the cure, and why on this one the CEO of Microsoft thinks the same.
I’ve just come back from WMF, the Web Marketing Festival in Bologna, 2026 edition, otherwise known as the big startup fair. I went in curious and came out with a question that has been going round my head for days.
There’s everything in there, and it’s good that there is: stands, pitches, founders with shining eyes, an impressive quantity of projects. It takes guts to get up on a stage and say “I built this”, and that kind of guts I genuinely respect. But the more I walked between the stands, the more an uncomfortable thought formed in my head, which is that what I was looking at, in the overwhelming majority of cases, were not startups. They were wrappers.
Wrappers, not startups
Let’s play an honest game and take them apart one by one. Nearly all of them are a thin layer built on top of an AI model that already exists, and each one covers a single vertical slice of a company’s operations: one does brand content, one SEO, one SEM, one social advertising, one handles expense reports, one the CRM, another measures employee satisfaction. Taken on its own, each of these solves one KPI, maybe two, at most three.
The consequence is fairly absurd: if tomorrow I genuinely wanted to run my company on these tools, I’d have to buy twenty-five of them, then thirty, then forty, then fifty, one SaaS per box, each with its own subscription, its own dashboard, its own login and its own invoice.
And I’m not throwing numbers around, because the average company already uses around a hundred of them. According to Gartner, moreover, nearly a third of that spend is “toxic”, made of dead licences, duplicated features and tools that do exactly the same thing: money walking out of the door on software left to gather dust. It’s no accident that when you ask people, plenty of them tell you the tools they use every day overlap with each other.
This is the world the wrappers are feeding, not curing. And in fact, if I have to be brutal, 50-60% of the “startups” I saw at WMF were not startups at all: they were ideas, prototypes, wrappers answering a market demand that was often non-existent, or nearly so.
AI was supposed to do exactly the opposite
The point is that I am deeply convinced AI exists to do the exact opposite of all this. It should let a company build its own software in-house, and where it isn’t ready yet, it should push the company to bring in the right people to do it, building the management system it actually needs, cut to fit its own processes, instead of buying it in pieces from twenty different vendors.
And this isn’t some obsession I picked up as a trade-fair veteran. Satya Nadella said it, the CEO of Microsoft, which is to say one of the companies that practically invented SaaS: on the BG2 podcast in late 2024 he explained that in the era of AI agents business applications are destined to “collapse”, because at bottom they’re nothing more than databases with a bit of logic stuck on top, and that logic is all moving inside the AI. Put plainly, the value no longer lives in yet another app, but in whoever governs the intelligence running inside it.
So what actually happened? What happened is that AI multiplied by ten the number of companies selling SaaS. There were already too many before, now there are many more, all with their own agent, all with a “powered by AI” of varying depth, and all busy covering their own little slice of the world. The technology that was supposed to unlock innovation has in practice flattened it, and to me that looks like a gigantic missed opportunity.
Don’t get me wrong: the problem isn’t the wrapper, it’s the wrapper with nothing underneath
Be careful here, though, because it’s extremely easy to overstate this, and I’m not arguing that every wrapper is rubbish. Cursor, to take the most spectacular example, was born leaning on OpenAI and Anthropic models (technically a fork of VS Code built on top of third-party models, rather than a wrapper in the strict sense) and became one of the fastest-growing software products in recent history, up to the roughly 2.6 billion in annualised revenue that in June 2026 earned it an acquisition agreement from SpaceX for 60 billion. Harvey in legal and Abridge in healthcare tell exactly the same story.
The difference, then, isn’t “AI yes or AI no”: it’s a single word, moat. Proprietary data, a vertical workflow the base model doesn’t hand you for free, real switching costs. The wrappers that die are the ones with nothing underneath, and according to estimates from people working inside this we’re talking about roughly 95% of them, the ones that stop at the first layer and end up fighting each other on price alone. It’s no accident that CB Insights and Gartner projections suggest 80% of AI startups could disappear by the end of 2026. Value, in short, doesn’t evaporate: it moves, and it ends up in the hands of whoever kept that data and that workflow in-house.
Who will actually win (and it won’t be whoever “added AI”)
The companies that will eat this market, then, won’t be the ones with the prettiest-looking wrapper, but the ones that have integrated AI inside and develop it themselves. There are already some signals in that direction, even if the road is a good deal bumpier than the headlines suggest.
Klarna, for instance, publicly dropped Salesforce and Workday, stating it wants to “switch off” several SaaS vendors as it consolidates, and in the meantime saw revenue per employee climb from around 575 thousand to nearly a million dollars in a year. It should be said honestly that it didn’t replace them with a single in-house AI, but with a mix of other tools and internal solutions, and that on another front, customer service, it actually reversed course and rehired staff: the direction is right, in other words, but the execution is still a work in progress.
Chamath Palihapitiya, for his part, launched 8090, a sort of AI-based “software factory”, and says one of his clients is already using it to retire a SaaS vendor costing 15 million dollars a year, rebuilding the solution internally at a fraction of the cost.
The common thread, in all these cases, is always the same: using AI to build the software that fits exactly you, customised, able to change along with the company and governed by the company itself, without a third-party IT department, a third-party consultancy and a third-party piece of software for every single function. And it’s almost trivial to see why they win in the long run, because all it takes is for new companies to appear doing identical things with AI embedded from day one, and against them the price war is lost before it starts. They’ll cost a hundredth, or even less, on OPEX, processes, services and quality, and the hundredth isn’t a provocation but an industrial plan somebody is already trying to put on the ground. The competitive advantage, in the coming years, won’t be having AI: it will be knowing how to build it in.
One thing, though, I liked
I don’t want to be the person who criticises other people’s work comfortably seated at a fair café, because I did see a couple of good things. The first is the startups working on space, engineering and robotics: nice, interesting, though in my view too far ahead of their time, and startups that are too far ahead are nearly always the ones that die first, not the ones that win. At least, though, they’re having a go at something real.
The second thing I genuinely liked: AI has allowed a lot of entrepreneurs who aren’t developers, engineers perhaps, or people coming from pure business, to build their own prototype themselves. Even when the underlying idea is wrong, even when it’s obvious it isn’t going anywhere, the fact that today you can turn an intuition into something tangible without a team of developers behind you remains an enormous change. And that one, yes, is deeply positive.
So?
In the end the problem isn’t AI, it’s the way we’re using it. We’re filling the market with tools that make one little piece more efficient at a time, when we could be building companies capable of redesigning themselves from the inside. It’s the difference between buying fifty SaaS products and becoming the company that no longer needs those fifty.
Let me be clear, and I say this for honesty’s sake: the “death of SaaS” is still very much an open debate, and today there are genuinely very few companies building their software in-house. But the direction, as I see it, is set by now, and whoever starts today, while the others are busy buying their fifty-first subscription, will find themselves holding years of advantage.
And here, to be honest, is the most important thing I’m taking home from this fair, far bigger than my irritation at yet another wrapper. I enrolled at university at forty-one, after founding and selling a few companies, simply because I’d realised I still didn’t know enough. And what I see today is that, for the first time in my life, the tools for building something real are no longer the privilege of a few, they’re within reach of anyone willing to roll up their sleeves and learn. The barrier that for twenty years separated the person with an idea from the person who can actually build it is crumbling, and to me that looks like the most valuable gift this technology is giving us.
Which is why I don’t read this moment as a sentence, but as an invitation. The historical moment we’re living in rewards those who build and not those who accumulate, and the truth is that the things that count were never brought into the world by people assembling parts bought elsewhere: they were built by people with the courage to get their hands into the engine. It has never been this easy to have a go, and perhaps it has never been this good either.
