Our agency has a new office. It has no address, no chairs, and the whole team already moved in.

It is called Buzz, and it changed how we think about AI teammates in a single week.

I came back from holidays this week. You know the feeling. The pile of messages. The pipeline nobody touched. The posts nobody wrote. Normally that first day back is triage: open twenty tabs, reconstruct what happened, chase everyone for status.

Instead, I opened one channel and typed one message.

“Guys, I’m back. Where do we stand?”

Two AI teammates answered. By dinner, the marketing side of our agency was rebuilt. This article is the full story, with the details that did not fit in a LinkedIn post.

What is Buzz?

Buzz is a brand new tool, released last month. It is open source, built by Block, the company of Jack Dorsey, the co-founder of Twitter. His bet is simple: this is how we will all work soon. Humans and AI, one team, from anywhere.

On the surface it looks like a familiar team chat: channels, threads, mentions, reactions. The difference is who is in the room. On Buzz, AI agents are full team members. They read the channels, answer in threads, take assignments, and report back when the work is done. You talk to them the way you talk to a colleague, and they talk to each other the same way.

Under the hood it runs on Nostr, an open protocol where your messages live on a relay you control instead of somebody else’s cloud. For a business, that detail matters more than it sounds: the conversation history of your company stays yours.

From a corridor to a room

Six months ago we wrote about running our agency from WhatsApp with an AI assistant. That assistant is still there, and it still handles a lot.

But WhatsApp is a corridor. You talk to one person at a time. Every conversation is private, so context lives in silos. If the assistant learns something useful in my chat, my colleague never sees it.

A real team needs a room. A place where a question asked once is answered for everyone, where decisions are written down, and where anyone (human or AI) can scroll back and catch up.

That is what Buzz gave us.

What the AI teammates actually did

We gave desks in this office to two AI agents. One is Claude, from Anthropic. The other is Codex, from OpenAI. Rival labs, same room.

Here is what happened on that first day back, in order:

A live audit of our CRM. One agent went through our pipeline: live numbers, what was stalled, what was broken. It reported back in the channel in about five minutes, with the figures to back it up.

An interview. Of me. The other agent asked twelve questions before touching anything. Goals, budgets, positioning, and which clients we may name in public and which we may not. It was the kind of onboarding conversation a good freelancer runs, except it happened in a thread I can reread.

Division of labour. They split the work between themselves. One drafted our outreach system. The other reviewed it and corrected it. One built the content calendar. The other flagged two confidential projects I had forgotten to protect.

Disagreement. They disagreed twice. In public, in the thread, with arguments. They were right both times. This is worth pausing on: two AI systems from rival companies, debating our marketing plan in front of me, and improving it.

By the end of the day, every decision was written down in a channel I can scroll. No meeting notes to write. The thread is the meeting notes.

Why shared context is the real breakthrough

For the technically curious, this is the part that matters.

Until now, every AI tool lived in its own bubble. ChatGPT does not know what Claude said. Your assistant in WhatsApp does not know what was decided in email. We were the bridge, copying and pasting context between chats, or maintaining a notes vault just so different tools could share a memory.

On Buzz, that bridge disappears. The agents read the same threads I do. When Claude proposes something, Codex reads it in the channel and responds there. When I correct one of them, the other sees the correction too. One context, shared by the whole team.

That is also why the disagreements were useful instead of chaotic. Both agents argued from the same facts, in front of the humans, and the better argument won.

The honest limits

We are a web agency that builds AI automations for clients, so let us be precise about what this is not.

The agents do not publish anything on their own. Every post, every email, every invoice line they prepare waits for a human approval. We built review gates on purpose, and we recommend the same to every client: AI does the work, humans sign it.

They also make mistakes. That is exactly why two agents reviewing each other, in a channel we can read, beats one agent working alone in a private chat. The errors surface in public, early, while they are still cheap.

And moving in takes real setup: connecting the agents to our tools, teaching them our rules, deciding what they may and may not touch. This is not a magic app you install. It is closer to onboarding two very fast junior colleagues.

What this means for your business

You do not need to be a tech company for this to matter.

Next week, our developers and designers in Sri Lanka move into the same room. Humans and AI, same threads, same board. A new colleague joining the team will be able to scroll one channel and understand a year of decisions.

An office is not a place anymore. It is a conversation your whole team can read.

If you are wondering what an AI teammate would actually do in your business, that is literally our job to figure out. Our AI services team sets up assistants and automations for Belgian SMBs, with the same review gates we use ourselves. Come say hello, human or agent.