Juicy news in artificial intelligence: Anthropic has blocked access to Claude from Windsurf.
Who’s who
If you’ve never heard these names, briefly: Anthropic is a competitor of OpenAI, so we’re in the world of artificial intelligence. OpenAI is essentially an arm of Microsoft; Anthropic is essentially an arm of Amazon and a piece of Google. Either way, they’re the new protagonists of AI, and they offer their artificial intelligence as a service so others can build products on top of it.
There are lots of those products - Lovable, a thousand other services, even Perplexity in the beginning. Under the hood, as their language engine, they use OpenAI, Anthropic and so on. Claude is Anthropic’s AI product, so we’re talking about Amazon.
Recently a series of unicorn startups were born around code editors for developers. Developers, when writing their programs, building sites, applications or whatever, use modern tools where artificial intelligence helps them write those applications and programs.
What happened? Essentially two big ones emerged: one is called Cursor and the other is Windsurf, both American. OpenAI tried to buy Cursor, the bigger one, didn’t manage it, and went for Windsurf instead, for something like 3 billion if I remember correctly - figures we can only look at through binoculars around here.
Windsurf is a code editor, a tool for programmers, which relied mostly on Claude. Long story short: Anthropic has now blocked access to Claude from Windsurf, with the excuse - a legitimate one, too - that they can’t provide their service inside products that became OpenAI’s through an acquisition.
Why should we care?
And you’ll say: “Piero, why should we care about these squabbles between startups and capitalists? We’re doing just fine.” For example, today I came to the gym, folks: six months of membership and I’ve been in twice, so every visit cost me hundreds of euros. But what a workout.
So what’s the point? The moral these fights teach us is that you can’t trust these APIs, you can’t trust these services. The access to artificial intelligence you take for granted - delivered through the cloud, which is the bulk of AI, so you can put it into three thousand other applications and products - can be wrecked, taken away or broken from one moment to the next, depending on who you do business with and who you get into a relationship with.
This is the extreme case of a full acquisition by their main competitor, but once again it confirms that the cloud is someone else’s computer, and if that someone disagrees with what you’re doing, you pay dearly. It’s the usual cloud dynamics, now hitting artificial intelligence.
What can we do?
The further conclusion: what can we do? Local artificial intelligence, private AI, data centers and cloud providers - maybe Italian ones - that you can trust. I always mention Seeweb, not because they ask me to, but because they’ve genuinely helped me out over the last year. Seeweb is an all-Italian provider: you get GPUs, you get LLMs as a service. But there are many other options. Can you trust these American hyperscalers doing crazy things?
Then of course, the problem is: who will give me Claude to run locally? Nobody. Claude is closed, it’s theirs, period. Are open models up to the level of the biggest ones? Yes, if you take the large ones - Llama 70B and up, and all these models in the tens of billions of parameters - they’re genuinely powerful.
But when you go into our business fabric, especially small businesses that maybe want private AI because they understand these dynamics and handle sensitive data, you have to put them in front of significant hardware costs. And they have expectations of a certain level of intelligence - rightly so. Then you give them something that looks like the same artificial intelligence they use online, except it seems to have taken twenty pizzas to the face.