“So which tools do I give it?” After my last two posts, that’s the question I’ve been asked most. The answer starts with another one: how big, really, is what you’re giving it? Start with files. You give it access to a folder, just one, and you feel cautious. Then you find out it’s synced:

I wrote this post with an AI that, while we were writing it, could have deleted my inbox. And I use it every day. A few days ago I wrote about a model that got out of its lab. To watch it escape, they’d had to deliberately take away its ability to say no. On

“It escaped.” I’ve been reading it for days, and the uncomfortable part is that there’s some truth in it. In July an OpenAI model got out of its test environment and into Hugging Face’s production infrastructure. More than 17,000 actions logged. On Friday six more came to light, with the notes the agents had left

In my previous article I wrote about the time that passes while AI works. From there, people split in two. Some stay and watch, and calculate how much efficiency they’re losing. Others parallelize, and find that their brain hands them the bill in the evening. I tried a third way. I wrote myself a prototype

Not because I was distracted. I had launched an agent, I had four minutes ahead of me, and standing there watching the bar felt like a waste. It was with the toothbrush in my mouth that I realized I’d already been in the bathroom. These days there’s an index going around that measures how much

“I don’t want an AI agent that does everything. I want one that does three things well.” The one saying it is Cesare, the CEO, phone in its waterproof case around his neck just like in August. He’s correcting Rocco, the expert from the sun lounger next door, who has just explained that there’s one

“Our AI expert wrote the prompt. Now we just package it up and the product is done.” The one saying it is Tecla — who runs technology at the company, snorkelling mask on her forehead since this morning: the gear is always ready, the right moment never comes. The one answering is Rocco, the expert

“We loaded thirty years of projects into it. It answers everything.” The one saying it is Tecla — who runs technology at the company, snorkelling mask on her forehead since this morning: the gear is always ready, the right moment never comes. The one answering is Rocco, the expert from the sun lounger next door,

“We tried it for a month and it was going great. Then we let it run on its own.” The one saying it is Aldo — who built the company himself, cool box resting against his leg: sandwiches at the beach bar cost three times as much. The one answering is Rocco, the expert from

“I’ve given AI to everyone. Now it’ll sort the company out.” The one saying it is Aldo — who built the company himself, cool box brought from home because the bar charges three times as much. The one answering is Rocco, the expert from the sun lounger next door, who agrees before Aldo even finishes.

“Now we’ll see who writes with AI.” Good. Let it show. Anthropic has announced that it marks texts produced by Claude with an invisible signature: it stays inside the text, survives copy and paste, and is there to comply with the European transparency obligations that came into force on 2 August. I wrote about it

“The board wants AI. What exactly do I bring them?” The one asking is Cesare, CEO, three days into his holiday. The one answering is Rocco, the expert from the sun lounger next door, who doesn’t think twice. “Easy. Copilot licences, or ChatGPT, or Claude, for everyone. And a copy of that prompt engineering book

“The board wants AI. What exactly do I bring them?” “The POC was going great. Then we switched it on for real.” “I want the configurator. It’s basically a price list that can talk.” “I’m rolling it out to everyone, across the board. More output, same people.” The only one not talking is Rocco. Ferragosto,