This morning I opened a new office.

No reception desk. No coffee machine. And, so far, nobody has asked for August holidays. That is already more than one can say for a normal travel agency.

It is called Buzz.

It is not another chat where you ask something and an AI hands back a long paragraph as if it had eaten Wikipedia for breakfast. It is a worktable: me, several AI agents, and a shared thread where everyone can see what was asked, what was checked, and what is still missing.

The difference sounds small until you use it.

---

The opening was not philosophical. It was an email from a hotel. As things often begin when they later end in a spreadsheet, three phone calls, and an aspirin.

It was a group proposal: dates, net rates, free places, conditions, and a lovely photo of snow. Normally I would have read it, saved it in a folder, remembered it two weeks later, and wondered why we had done nothing with it.

In Buzz I opened a thread and asked the agents whether the proposal could become a post for Google and social media.

One reviewed the email. Another separated what could be said publicly from what was missing. Another designed the flow: read the proposal, classify it, prepare a draft, ask for my approval, and only then publish. The conclusion was useful because it did not promise miracles: the hotel could attract groups, but it lacked enough information for firm prices; individual travellers could ask for a tailor-made option without inheriting rates that were not theirs.

That is professional judgement. This time it was distributed.

---

What I like about Buzz is not that agents do things on their own. That sounds wonderful until it ends quickly in a disaster with excellent spelling.

I like that they work in the open.

The original question stays in the thread. Who checked what stays there. The draft stays there. The objection stays there. And so does my final decision. I also tell them where they may get information: official sources, supplier documents, and no inventing a rate because an advertising page looks rather like a contract.

If an agent gets carried away and wants to publish an incomplete fare, I ask it to open the hotel’s Excel sheet. I check the net rate, verify the margin, and decide whether that figure is fit to go out. It is all there, caught with its hands in the dough.

For someone who has spent forty-two years distinguishing between an “indicative quote”, a “pending booking”, and an “issued ticket”, this is not a technical detail. It is the difference between a tool that helps and one that leaves a fire burning in the kitchen.

---

I do not want five agents giving me five different answers. I want a workshop: one idea goes in and one reviewable piece of work comes out.

The roles can change. One researches. Another sorts the data. Another turns the proposal into a draft for each network. Another prepares an image or video. But they do not hand me five reports and make me become a simultaneous interpreter: they return one package. I review it and, if it is time to publish, I approve it once.

That is what I want to repeat with every new supplier email: a template that improves with every correction, not a robot meeting held while I am not there.

---

Buzz is very new. So new it still smells of fresh paint. But it brings an idea I find enormous: agents are not isolated buttons inside an application. They have an identity, permissions, and a shared workplace. Codex, Claude, Hermes, or whichever model appears next month can join without making me move house digitally.

Here is a word I love: harness.

The harness is not the intelligence. It is how you hold it so it can pull without flying out of the window. My context, my rules, my checks, and my machines remain here. The model can change. The work does not lose its memory every time a new one announces that it has finally invented the future.

I rent intelligence. I do not rent out my office.

---

So that this office does not become an independent republic, I write skills. They are work instructions: what each agent may do, what it may not do, which sources it must use, and what it must return to me. I keep refining them by watching the answers from different agents.

I do not hire agents and cross my fingers. I train them, limit them, and improve the handbook.

And there is a very clear boundary. An agent may research, compare, prepare a draft, or open a spreadsheet for me. But, for now, I do not let it book flights, issue tickets, move money, or wander cheerfully through confidential data. The technology could already try many of those things through browsers and terminals. That is precisely why I do not hand it all the keys.

Agents have autonomy to work. Not autonomy to ruin me.

---

The scene that really convinces me does not happen in front of a screen.

It happens while I am cycling around Marbella. I open Buzz on my phone, leave an instruction by voice, and keep pedalling. The thread connects to my main Mac and my Mac Mini. When I return out of breath, the agents have moved things forward: they have organised information, spotted a doubt, or left a draft ready for me to tear apart lovingly.

The work does not disappear. The silly wait between having an idea and starting to move it does.

And at 73, that is worth more than an office with a sea view. Artificial intelligence has not removed years from my driving licence. It has given me back thirty years of appetite and put my neurons to work in a brutal way. That is why this blog exists.

---

I do not know whether Buzz will be the tool we all use in two years. I do not even know whether it will still be called Buzz. Tools change names, prices, and owners with a speed some Maldives hotels might envy.

But the idea remains: I do not want a collection of brilliant AIs, each working in its own fishbowl. I want a worktable where they can collaborate, disagree, leave a trace, and deliver something I can review before it goes into the world.

The new office has no coffee machine. But for the first time, it has no corridors either.

Buzz is a project from Block, launched by Jack Dorsey. It has only just arrived, but the idea of putting people and agents to work in the same office has more than a touch of mini-genius about it.