Most agencies sell you a deck and a meeting. We are going to do the opposite. We are going to open the machine and let you watch it run, the whole studio, operated by AI, in public, minus the hype. Here is the honest version of how this works. A studio is not a mystery. It is a set of repeatable moves, and once you can name each move, a model can make it for you. So for the next few minutes I am going to walk you through the actual pipeline we use to plan, write, render, and ship content, the same one that built the video you are watching right now. No secrets withheld. By the end you will be able to copy the structure yourself, because the structure is the only thing that matters.

It starts in the terminal. The brain is Claude Code, running as a coding agent, holding the entire project in context instead of you clicking through a dozen browser tabs. We used to run flows in n8n, dragging nodes between boxes. We did not throw those away, we transformed them, taking the n8n JSON and rewriting it as typed Pydantic Python, where the types are the contract and a test can prove the flow before it runs. Now the model calls every tool directly through what are called MCP servers, small adapters that let one agent talk to ElevenLabs for voice, to HeyGen for an avatar of the operator, to Remotion for video, to Airtable for records, to a browser, to the file system. The move you can steal is this. Stop thinking of automation as a chart of nodes. Think of it as one operator with a memory and a toolbelt. You describe the job in plain English, the operator reads the project, picks the tools, and does the work. That single shift, from clicking to instructing, is what makes a one-person studio behave like ten.

Here is the part people miss. Our workflows are not code locked in an app. They are data, plain files on disk. The entire content calendar lives in one file we call the content index, a big structured document where every single asset, every video, every thread, every long-form like this one, is a row with a pillar, a date, a duration, and a list of scenes. When we want a video, the agent does not improvise. It opens that file, finds the asset by its identifier, reads the scenes, and works from the spec. Want to change the plan? You edit the file, not the machine. This is the give-away. Put your strategy in a flat file the model can read, anchor every item to a date and a topic, and your calendar becomes something an agent can execute instead of something you have to remember.

Now watch what that buys you. One brief, many outputs. Because the plan is data, a single topic does not become one post. The agent reads the source idea and writes it through different lenses for different surfaces, a sixty-second short, a ninety-second explainer, a long-form like this, a thread of several scenes for the feeds. Same spine, different shape, sized to the platform. The old way was a writer producing one thing, then a second person reformatting it five times by hand and losing the thread each time. The method now is to hold one source of truth and let the model project it into every format in a single pass. You write the argument once. The machine handles the translation into nine outputs, and they all still say the same true thing, because they all came from the same row in the same file.

But generated work without checks is just noise at scale, so here is the spine of the whole thing. Gates, not vibes. Before anything ships, it runs through automated checks we call gates, plain scripts that pass or fail. One gate verifies the brand never breaks its own rules, like the script that refuses any banned hype word or a logo where it should not be. Another is a parity gate that confirms the back-end plan and the public page actually match, so what we promise is what we built. A variety gate makes sure ten videos do not all look the same. Nothing reaches an audience on a feeling. It reaches an audience because a check returned green. Steal this directly. Write down your standards as code that can fail loudly, then never ship around a red gate. That is the difference between a studio and a content treadmill.

Then comes the render, and the render is just a function. The video you are watching is not filmed. It is React code, drawn frame by frame by Remotion, a library that turns components into video. The moving shapes behind me are React-three-fiber, what we call R3F, drawing our Signal Field in real three-D, the same scene that runs live on the site; for visuals that have to react in the moment we drive TouchDesigner, generative graphics controlled like code instead of rendered once and frozen. The scenes you saw in the content index get passed in as properties, and the same composition renders every asset in the calendar, just with different inputs. Voice is generated separately through ElevenLabs from the same script you are hearing, then muxed onto the frames. So one composition, fed by data, produces hundreds of distinct videos, each pinned to an exact version so it renders the same way every time. The lesson is to treat your output like software. Define the look once as a function of the content, version it, and let identical inputs always produce identical results. Repeatable beats clever, every single time.

And then we publish the receipt. We run an office, a live public page on temerarii dot xyz, where the work and the method are simply on display, the calendar, the system, the pieces themselves. This is the build-in-public thesis stated plainly. We do not tell you we are good. We show you the running machine and let you check it. The old agency move was a polished case study about a client you cannot reach. Our move is the live page, the actual file, the nine outputs from one brief sitting right there. Receipts over rhetoric. If you want people to trust how you work, the strongest thing you can do is stop describing it and start exposing it, the calendar, the gates, the renders, all of it, in public, where anyone can audit the claim.

Now, none of this means the humans disappear. It means the humans move to where they actually count. The agent does the mechanical labor, reading the file, calling the tools, rendering the frames, running the gates. The human does taste and judgment, choosing the topic that is worth saying, writing the argument that has an edge, deciding the gate is too lax and tightening it, looking at a render and saying that one is dishonest, cut it. The model is fast and tireless and has no opinion. You bring the opinion. The old way burned your best people on formatting and uploads. The method now is to hand every repeatable move to the machine and reserve every human hour for the decisions a machine should never be trusted to make. That is the whole trade, and it is a good one.

So do not take my word for any of it. Open it and check. Go to the office, find this exact long-form in the calendar, and read its row, the pillar, the date, the ten scenes. The script on that page is the script you are hearing right now, because there is one source and one track. Look at the other assets from the same week and you will see the same topic projected into a short, a thread, an explainer. That is the one-brief-many-outputs claim, verifiable, not asserted. Everything I just described, the terminal brain, the data calendar, the gates, the render function, the public receipt, leaves a trace you can inspect. That is the test of build-in-public. If the method is real, the evidence is sitting in the open, and the proof is that you can find it yourself.

So here is what is now possible, and it is simpler than the pitch usually makes it sound. A studio is a set of named moves. Put the plan in a file, give one agent the tools, write your standards as gates that can fail, render from data, and publish the machine instead of a brochure. That is the whole method, given away, because the structure was never the secret. The discipline of running it is. And if you would rather not wire it yourself, this same engine is a product you can run, an app we call SlotSlot, the command center where the calendar, the tools, and the gates are already assembled. Everything we make goes through this, and everything we make is on temerarii dot xyz, in public, where you can take it apart. Go build your own version. We would rather show you the machine than sell you the meeting. That is the studio. That is how it runs. Now you have got it too.