This week the hub goes from a promise to a button. We are going to walk through the one move that runs the whole place: you pick a workflow, you press run, and a chain of tools does the work while you watch. By the end you will know how that chain is wired, in plain steps, with the real tools named. No theater. The Big T-M runs this in public so you can copy it.

First, drop the mystery. A workflow is just a list of moves written down, in order. Open a file called run dot yaml. Each line is one step: fetch this, write that, post there. The model reads the list top to bottom. If you can describe a job to a coworker in five sentences, you can write the list.

The runner is Claude Code in the terminal. It is one coding agent that keeps the entire job in its head instead of you clicking between tabs. You hand it the list and say go. It does not forget step two by the time it reaches step six, which is the part humans get wrong at four in the afternoon.

The agent needs hands to touch real tools. That is MCP, the Model Context Protocol. Each tool gets a small adapter, an MCP server, that the agent can call directly. One server for your sheet, one for your inbox, one for the video renderer. You add a server, the agent gets a new hand.

Step one of most runs is reading data you already own. Wire an MCP server into Google Sheets or Airtable, point it at one tab, and let the agent pull the rows itself. Now the run starts from your real numbers, not from a guess the model made up.

Once the data is in hand, the agent passes it to the next step on the list. It calls the writing step, then the render step, then the post step, each one a separate tool through its own server. You wrote the order. The agent walks it. You are reading email while it happens.

Never let a run touch the outside world without a gate. Put a line in the list that says stop and show me. The agent pauses, prints what it is about to send, and waits. You type yes. This one habit is the difference between a tool that helps and a tool that embarrasses you in public.

The payoff is that the list is reusable. The same run dot yaml works tomorrow, next week, with new rows, with no new clicking. You changed a job from a thing a person does each morning into a thing a machine repeats on command. That is the whole point of the hub.

Here is the honest proof. The Big T-M does not just talk about this. Our whole content calendar is built from one source file by the same kind of run, and the office site you can visit is the output. We eat what we cook, in the open, where you can check it.

So that is the move: write the list, point one agent at it through MCP, gate the risky step, press run. Start with one boring job you do every week. Go see the live walkthrough at office dot temerarii dot xyz and watch a real run from the first line to the last.