Why I'm finally building Televism

A love that never let go

I’ve loved MadTV ever since I first played it at a friend’s place on his PC. The trouble was: it wouldn’t run on my sister’s 286. We had nothing better in the house. So I started delivering newspapers to save up for my first own PC. Anyone who has ever carried papers from mailbox to mailbox in all weathers for a goal knows the feeling: something you’ve worked that hard for doesn’t just let you go again.

For me that still holds true today. Sure, at some point MadTV was pushed aside by more modern games. But a quick round in DOSBox still works.

When I started university, I helped write a remake with a team scattered across the internet — tvtower.de. A shout-out to Ronny at this point: I admire your dedication and stamina. You’ve been driving this project forward for nearly 25 years now. I wish you continued success!

My problem is called “and now just do it”

Here’s my honest blind spot: I’m a prototyper, not a productizer. I love the first phase — the idea, the big picture, figuring out whether and how something works. But once the honeymoon phase is over, the picture clearly drawn and really all that’s left is to “do it,” I lose interest. Right at the moment a project becomes routine for the first time.

You can trace the idea of a MadTV remake like a thread through my entire technical biography — each time dressed in the fashion of whatever era I happened to be in:

When I first came across servlets and Hibernate: “Man, MadTV as a browser game — now that would be something!”
First big project at a corporation, and suddenly the explanation for why it never worked out: “Ah, I need a functional spec and a technical spec, and then I just have to work through it.”
Later: “Browser games are so over. I’ll do this as a mobile game, Unity looks like the perfect tool for it.”

Every time a new run-up, every time the same pattern. The beginning excites me, seeing it through gets left behind.

The old spec that still holds up

A few months ago I stumbled across my old functional spec for “MadTV as a browser game” again by chance. And my first thought was: Man, that was really good. A little dusty after fifteen years — but still good at its core. Why did I actually never build it?

Even the very first paragraph pulled me right back in (despite the wooden language):

Fun manifesto

The following principles are declared, to be upheld in order to ensure maximum fun:

Bots are expressly allowed! If a game has tasks that can be done better automatically than a player can do them, there is a flaw in the game design.

Every player, no matter when they join the game, should have the chance to be among the top players. No advantages may arise from long playing time.

The goal is to achieve a maximum of complexity with a minimum of game rules.

The nature of the rules should be understandable and memorable through common sense or by transferring real-life experience.

The complexity of the game should increase with growing success.

Maximum success can only be achieved through the interaction of several players in a group.

These six sentences have not aged badly over the years. They are still the compass Televism aligns itself with.

“Just feed that into an AI”

The obvious thought today: throw the document into an AI and let it do the work. Of course the first attempt produced nothing but rubbish — what else would you expect when you dump a 20-page Word document full of functional descriptions, diagrams and screenshots into it and say “go”?

But my interest was sparked. Not the old question (“How do I build this?”), but a new one: How do I get an AI to implement my thoughts from fifteen years ago?

Show it once, then scale

And this is exactly where AI turned out to be a real turning point for me — for a reason that has little to do with “the AI writes code” and a lot to do with my own psychology.

With Televism I’ve already been through several cycles of “awesome new challenge!” and “…and now just do it.” I always loved the challenges. The “doing” — coding ten functions to the same pattern — was always the cliff that earlier attempts crashed against.

That very “doing” is the ideal task for a coding assistant. We develop a new function or a new design together once. Then we expand it into a pattern — and the assistant can implement the rest largely on its own. The pace at which it clears away the dull parts keeps me in the exciting part — and therefore keeps me motivated.

What’s coming

For me this project is like a puzzle — only one without a reference picture, where I don’t even know how many pieces it will consist of in the end. But a few outlines are already taking shape:

Bots that balance the game rather than break it.
Cheap, privately fundable infrastructure instead of enterprise Kubernetes in the cloud.
Station profiles and market-driven viewer interests
Repository structures in which an AI agent can find its way around optimally.

How all these pieces come together into a picture is what I want to report on here.