My exact setup: Claude Code, Buzz, and a team of AI agents

I run a video production company and I am building a software product on the side. I do not have an assistant, a project manager, or an ops person. I have a small team of AI agents, and they do most of that work.

This post is the exact setup, written plainly. It is the same setup I install for clients, so if you are wondering what working with me actually gets you, this is it.

The engine: Claude Code

Claude Code is Anthropic's AI that works on your computer, not in a browser tab. It can read your files, write and run code, search the web, and use your tools. ChatGPT answers questions. Claude Code does tasks.

That difference matters more than any model benchmark. An AI that can only talk to you is a clever advisor. An AI that can open your calendar, read your inbox, and update your task list is an employee.

Where they live: Buzz

The agents work inside Buzz, a messaging platform that looks a lot like Slack, except AI agents are members alongside the humans. Every project gets a channel. I message an agent the way I would message a person, and it goes away, does the work, and reports back in the thread.

Each channel gets its own session of the agent, so the ClawUK channel knows about ClawUK and the video company channel knows about the video company. Context stays where it belongs.

The team

My agents have names and jobs, the same as any team:

Max is my right hand. Planning, research, drafts, second opinions, day to day thinking. John handles outreach. Trevor handles content strategy. Eric will handle ads when we switch them on.

They share a workspace, a memory of how I like things done, and my task list in Linear. When I wake up, there is a digest of what got done overnight, what is blocked, and what needs a decision from me. I make the decisions. They do the doing.

What a normal day looks like

Morning: I read the digest of what got done overnight.

During the day: I dictate messages between jobs. "Max, look at this ticket and tell me what you think." "Draft the follow up for that lead." The agents pick it up, do it, and post the result where I can see it.

Evening: I review anything outward facing before it goes. The agents draft, I approve. That division of labour is the whole point. My judgement, their hours.

What it took to set up

Honestly? Months of tinkering, breaking things, and learning where the sharp edges are. Which credentials to give an agent and which to never give it. How to write instructions it will actually follow. What to automate and what to keep human.

That learning curve is the reason ClawUK exists. You do not need to spend the months. I already did. In a consultation we map what your version of this looks like, and if it makes sense, I build it for you, on your systems, with your team trained on it.

If you are curious where to start, start with an hour. Worst case you leave with a plan.