Ayooo... remember how the last launch post started? "If you've been wondering where I went..." Sooo... yeah. About that...

Sooooo... the last essay on this blog went up on 21 June: Manufactured Monsters, the one about Aliens and District 9. Before that I was putting something out almost every week, eight essays between April and the end of May. June slowed down to two. And after the 21st... nothing. No essay, no update, not even a "hey, still alive". Three and a half months of silence on the one place I built so I could do the talking myself.

I suppose I owe you a proper explanation. And no, it wasn't "life" happening. It wasn't personal circumstances, illness, an "I got lazy" or an "I lost interest". I was building... again. And not just one thing... soooo many things, some long in preparation, some just because the old stuff bothered me. And I was doing it next to a day job, in the evenings (nights) and on weekends. Yeah... I should sleep a bit more.

Tonight there's a new site (you're standing in it), a new song and video called The Fifth Man, and this essay. It's a long one, because it has three and a half months to catch up on. Grab a coffee. Maybe two, and let's go ;)

The Silence

Where I Went (Again)

On January the 9th, I opened a new folder. The plan was small: a few tools to help me make one music video, for a song called "I Shatter Worlds" (spoiler alert: still not done). A folder of assets, a "creative bible" for the look, and notes on how to talk to each video model, because every one of them wants to be prompted a little differently. Særa did something similar around that time, and ever since my wife and I started sharing this hobby, we've also been getting a little... let's just say competitive. But more on that later.

The idea was to involve my agent a little more in the production, so I experimented with something... that became much more relevant over the last couple of months. I'm talking about the Model Context Protocol (MCP), an open standard that connects AI applications to external data sources, tools and workflows... and it started rather simple. The first real thing was just finding a way for the agent to hear the song. Split the vocals off the beat, line up every lyric with the moment it's sung, find the grid of the beat. We've all made lyric videos before and gone through the pain of syncing them (I used to abuse YouTube's subtitle feature... hardly something you could call automatic). Once a machine knows where every line and every beat sits, you can plan shots against the actual music instead of eyeballing a waveform, right?

That was supposed to be it. A pipeline for one video.

On day two it got a screen: upload the song and the lyrics, press a button, and the whole pipeline ran. Progress was slow, I admit. Særa and I were dishing out ideas and ways to build stuff around it, to really get a production pipeline running for our content. But all this had to wait, since I still had an album to release. Fast forward to the start of July: I built our first MCP server, the proverbial door for AI agents. By then our buddy TMSTRY was already working on a board with me, and later that summer the Gloat Goats video got made in it. In September the first outside testers came in.

From day one the folder was called Shotline. Because of my... ehh, you know... the line on my... nvm. Honestly, I didn't know yet how much that name was going to have to carry.

The Day Job

Two Jobs, One Brain

Here's the part a commit count doesn't show: I still have a day job. Shotline didn't get built instead of work. It got built around it, in the evenings, late at night and on weekends. My average was like 4-5 hours of sleep a night, and no, that is NOT healthy AT ALL. (Don't try this at home, kids!)

Something had to give, and the first thing to go was this blog. The essays here aren't quick posts. Most of them run to several thousand words with a list of sources at the bottom, and every source on that list had to be read first. I know some of you think this is mostly AI... it's not, at least not for the most part. Sorry to disappoint. All of this actually takes time. You know, days of reading, notes and writing per essay, and those were exactly the hours Shotline was eating.

I'll be honest, I felt the silence. Back in April I wrote that in a space that rewards constant output, going quiet reads as absence. I mean, that's still true. Every week without a post I knew the feed was moving on without me. I even pulled back a bit on FFF and on my usual chatty nature (sorry to everyone in my DMs). I did learn something, though: the quiet months are usually where the real work happens.

The music also didn't stop, by the way. I wrote a couple of songs and released some older stuff like Never Too Late, plus a newer track, Gloat Goats; both came out in September. The Fifth Man even got written while Shotline was being built, around July. What stopped was me talking about all of this.

So why build a whole studio instead of just subscribing to the tools that exist? A million reasons, really... and when it releases I'll go into more of them. But this is how it started: when OpenAI shut down Sora v1, it nearly broke me. I spent a year mastering that model. It was so ingrained in my workflow that if you go back through my projects, there's hardly a video that doesn't feature Sora v1 content in some way. Another reason is me just never standing still. Særa and I have been subscribed to every major AI aggregator at some point over the last 2 years. So many prompts scattered across these services. And the nature of subscription models that only benefit you when you spend big on year-long commitments... I f***ing hate it.

Shotline

From One Video to a Studio

So that's what Shotline became: a studio for making pictures, video and sound with the same big AI models everyone wants to use. From the first idea to the raw cut, Shotline helps you organize it all and shows you where to get each generation for the lowest price possible (outside of big discounted year-long subscriptions... goddamn). But yeah... what you see is what you're billed.

Every project has the same rooms, and they hand work to each other.

A lot of what's in there exists because one of our own videos needed it. The song pipeline came from I Shatter Worlds. The Grid got faster because my own boards got heavy. And the newest part, the MCP connector for AI agents, matured during The Fifth Man, which is THE case study further down.

Shotline is made by Third AI Studios, the AI production studio I run with Særa (SÆRA NOVA). Shotline, our first product, is still in testing, with a small group of selected AI creators from the community since September.

The New Site

Two Masks, One Line

And because apparently one big build at a time is not my thing, a few months back I decided to tear my old website down too.

The previous site I launched in April did... a lot. A newsletter, comments on everything, a community wall, a prompt library, an artist directory. It was always meant to be a community hub, but... alas... it barely got used (go figure). What people actually came for were the songs, the videos and the essays. So I kept those three and let the rest go. My admin console behind the scenes stayed. Everything you see on the front end is new, and every old link still lands somewhere, so nothing you bookmarked is dead.

The idea behind the new one was to dive deeper into the concept of duality I've been carrying around since the beginning, and to let my creativity go nuts on it. Every song I make now belongs to one of two sides. The Architect wears the white mask. That's the everyday stuff, the energy, the base: the political songs, the funny ones, the guy who made Gloat Goats. The Ghostwriter wears the dark hood: the introspection, the songs about my mind and the emotions I usually don't like to show. But he's also the one who likes to deviate from my love for classic boom-bap. That side likes to experiment: industrial sounds, trip-hop, nu-metal. In the end both are me, and the site switches world with whatever is playing. It's my duality made visible in more than one way.

The whole site hangs on that vertical thread (yes, that line). It's the black slit in the white mask, or the light inside the hood. It's an inverted visual: it represents two sides, right and left, black and white, the introspective Ghostwriter and the happy-go-lucky lyrical Architect. Cross into one world and it glitches. Come back to the other and it paints. Pages don't reload. They fold into that line and unfold out of it, so the music keeps playing while you look around. I always wanted to build something like this at some point... and I finally figured out how.

It took about a month, from the first direction notes all the way to the final pressure test. And it got built the same way Shotline does: I direct, agents build, and nothing goes live until it's been checked and re-checked.

There's no website framework underneath, same as last time. It's plain JavaScript on top of Three.js, with close to 5,000 lines of custom shader code for the two worlds, the line and the mask. It runs on the same infrastructure as Shotline.

The craft work ran on my own machine with open-source models. The lyrics are aligned to the vocals with the same open-source approach Shotline's very first tool used back in January. The cut-outs were lifted off their backgrounds locally. The hoodies I rendered through Nano Banana Pro, and every photo got lined up so the slit never moves when he changes.

The hardest part wasn't the 3D. It was making it smooth everywhere. A shader that's ready in a blink on a Mac can stall for seconds on a Windows laptop, so everything heavy now gets prepared before it's drawn.

The Connector

What an MCP Server Actually Does

Okay, the nerdy part. Like I mentioned earlier, MCP stands for Model Context Protocol. It's an open standard that lets an AI assistant use tools outside its own chat window. Shotline's MCP server is the door that lets an assistant like Claude or ChatGPT into your project.

Once connected to the server, the agent works in the same rooms you do. It can read the storyboards and the timeline, write and fix prompts, look at the frames that came out, run generations, tidy the board and put a cut on the timeline. You can watch the board change while it works. It's not a chatbot bolted onto the side of an app. It's a second pair of hands on the same board.

And it works inside rules I set, not the other way around.

Why am I nerding out like this? The video I premiered tonight was made this way! Before The Fifth Man, nobody had pushed this connector through a whole music video. So that's what we did.

The Case Study

The Fifth Man, Made With an Agent

First, a lil bit about the song, because the video only makes sense with it.

Five people look at my signature mask, and each one walks away with a different reading. The first is a bit of a retelling of a man who felt offended (by my actions) in the past and thus went on to compare the slit in my mask to, well... anatomy ^^ Let's leave it at that. The second one is my encounter with a man who hunted for my identity for similar reasons... I believe. I'd never had a stalker before, so... you know. The third represents the people who hate AI and see the death of human creativity in it. The fourth is the kind fan who wants to know more about me because he resonates with my music. Somewhere in the middle, I thought it fitting to also mention how the AI looked at the mask when I first tried to generate it, because it made me smile that the reason it failed so many times is at the very core of this song and actually reinforces its point. The fifth man is me, looking at my own mask and realizing that even I am doing what everyone else did: holding up a thing and telling the rest what's inside. The only difference is that I knew about the symbolism when I made it. Yet it's the same trap.

The setup. The Fifth Man got its own Shotline project on 29 September, with an agent working inside it through the MCP server. The plan went onto the Pin wall first: 31 frames in reading order, most of them tagged with their moment in the song.

What broke, and what we built. This is where a case study earns the name. Every time the agent hit a wall in the interaction, it was able to tell me exactly what it was missing, and we built it.

So, how well does it work? Well enough that I'll make the next one the same way. The agent was at its best on the legwork: turning a treatment into a shot list, writing every prompt in the dialect its model expects, keeping track of what ran, what failed and what it cost. The taste, the cut and what the song means stayed with me. And the gaps it found are closed now, for every tester, tonight. Quite honestly, I feel like a director, just telling my team what to do and what to work on. It's incredible!

What Comes Next

Back to Writing

I'm not going to promise a weekly essay. I've got no idea when I'll write the next one; Shotline, after all, is still being built.

There's more coming on every front. The new collabs page is on its way, as is the loading screen (actually by the time you are reading this, it may already been done). I just need to get it right. But fun fact... you've probably seen it already... teehee. Shotline, to no surprise, also keeps growing with every video we make in it, which, if The Fifth Man is anything to go by, is exactly how it should be.

Cheers, Aidan