No. 28AI Technology

Everyone's a director now

Someone posted a game developer’s take on building with AI to the three.js subreddit this week. The post was fine. The replies were the interesting bit, because they split down the middle.

The first one under it said All the fun is gone from development. The next said the opposite. A developer who’d burnt out on endless admin screens and boilerplate had come back to coding after two and a half years away, and was enjoying it again because all the stupid, boring, repetitive work is Claude’s problem.

Then someone sneered that everyone is a technical project manager now, and an art director with fifteen years behind them answered. They’d done VFX, shaders, modelling, lighting, concept art and game design, and directing was the most fun of the lot, when you have a great team. Having AI is akin to having an amazing team, on all the time, scalable. If you have taste and good ideas, they said, this is the best time there’s ever been to make things. It just takes a mental shift.

I’ve been on the far side of that shift for a week, and I agree with them.

The code was never the joy

Fred Brooks, who ran one of IBM’s biggest software projects, made a list of why programming is fun. The first item was the sheer joy of making things. That’s how it was for me. I wrote good code for thirty years, but the code never gave me the joy. What came out the other end did: the page working, the thing looking the way I’d pictured it. Wanting the result is what gave me the patience to write all the code in between.

A game needs months of that patience. Hardly anyone writes an engine from scratch any more, but even with a modern one doing a lot of the work there’s an enormous amount of manual dog work between the idea and something that looks any good. I’d never made a game, and it wasn’t for lack of ideas. I knew I’d never have the time or the patience to finish one.

One evening

What finally kicked me into it was VIPER 11, a free fighter-jet game in the browser made by a developer called apek, with three.js. Its itch.io page says the code, graphics and sounds were all made with AI’s help, and it’s beautiful. I’d had the Nazca idea for a while, and seeing his was the kick up the backside I needed.

So last week I made one. Trazo is a little browser game where you fly a cartoon plane over the real Nazca Lines in Peru and paint in the giant figures by keeping the line under your plane. Going by the game’s change log, the first evening went like this.

Fig. 1 · The first evening

Two hours, from a written spec to a game with sound. Every step is a line from the change log. Twenty minutes after the plane first flew, an AI could steer it too.

I haven’t looked at a single line of its code, or at how the engine works. I described the engine I wanted, then flew it and said what felt wrong. Everything after that was direction. I described the characters with reference points I already knew, the toy plane and the big-headed pilot, so there was something to aim at. I researched how other flying games handle and said what I wanted this one to feel like. None of it was “make me a game about the Nazca Lines”. It was a lot of detail, most of it dictated on walks, the way I driftcast everything.

It isn’t finished, but it’s playable and, I think, a lot of fun. This week it was the top post on the Peru subreddit. The replies, in Spanish, said it was a nice way to pass the time and a good idea, and one person was bracing for the plane-crash jokes.

A team in a box

What I had was the art director’s team. One person with an idea now has an engine programmer, a modeller, an animator, a sound designer, a composer, a voice cast, a copywriter and someone to put it all online, on all the time.

Fig. 2 · Who did what on Trazo

Every job but one done by the team in the box. My part in each was an idea, a line of direction or a choice. Playing it, and saying what was wrong, was the only job that stayed all mine.

Brooks drew the line forty years ago. In No Silver Bullet (1986) he divided the difficulty of software into its essence, what the thing has to be, and its accidents, those difficulties which today attend its production but which are not inherent. His bet was that the hard part was deciding what to build, not the labour of building it. The accidents are what the team in the box has taken off me. The essence is still mine.

So game development, for me, has become something like 99 per cent creative direction and 1 per cent technical. That’s the mental shift the art director meant. The time I’d have spent building goes on flying it again and tweaking it, which is what I argued for last time: not ten games because I can, but one good one.

Why games feel it more

The first wave of this was web pages and apps, and nobody much mourned the admin screens. Games are different. A games studio is full of creative people who aren’t programmers: artists, animators, writers, composers, voice actors. Those are the jobs in my second figure, and one person with a good idea can now get a long way without any of them.

It has come at a hard time, too. In this year’s GDC survey of the games industry, more than one in four of the 2,300 people who answered had been laid off in the past two years. 52 per cent thought generative AI was having a negative effect on the industry, up from 18 per cent two years earlier. Just over a third were using it in their work.

So when someone says all the fun has gone, I don’t think they’re wrong about what they’ve lost. If the puzzle of the plumbing was your joy, something real has been taken. It just wasn’t mine.

I’d still want people in. With more to spend, a real artist would make Trazo better. But you don’t need anyone to build a prototype and get it playtested. Bring them in later.

The flood

The sceptic in the thread had a point. The results will speak for themselves, they said, and their games will look about as inspired as the flood of AI generated restaurant menus. The web is going to fill up with games, and lots of them won’t be very good.

But the art director’s line had a condition in it: directing is the most fun when you have a great team. Directing is a skill too, and most people have never done it. You have to know what good looks like, and describe it well enough for someone else to build it. That’s the difference between a game and a restaurant menu. Another reply put it more bluntly: it depends on you having good ideas, and knowing the fundamentals well enough to know which order things go in.

What’s great about the flood is who’s in it. People who had an idea and never had the means now get to direct it and produce it. I’m one of them. I’m a developer, but I’ve never thought of myself as a game developer, and a week ago I’d have felt a fraud saying I was one. Now I’m making a game the same way friends who’ve never written code are building their own little tools to organise themselves. And I’m completely confident in it, because I’m good at the part that’s left. I know where the eye should go, and I can describe a mechanic well enough for it to get built.

It doesn’t do the same thing twice

The code I wrote was deterministic. It did the same thing every time, and even the random bits in a game follow rules someone wrote. Characters said what they were scripted to say. AI doesn’t, and games are where that’s most fun to watch.

I like watching AI play games, so on that first evening I made Trazo flyable by one. An AI connects to the game, gets told where the plane is and where the line goes, and decides what to do next. Its runs go on an AI board, ranked by model. The next day I read an article about someone who’d built a sailing game with Claude Code and raced the big models against each other. Gemini won easily, while Claude stalled and timed out. A small model wandered off to a beach. Nobody has really found my board yet, but I’d love to see it fill up with models trying to beat each other’s times.

Then there’s the Halo companion. A YouTuber called chasm modded the original Halo so that your sidekick, a Spartan called Kevin, is run by AI, and shared how. You talk to him out loud. In chasm’s words, his nervous system runs on a small model called Kev, which makes the quick decisions, so Kevin reacts straight away, and a bigger, uncensored model does the thinking and the talking. He swears, and he’s genuinely funny. Nobody wrote his lines.

Bigger studios are trying the same thing more carefully. Krafton’s Sims-style inZOI gives its characters a small AI model that runs on your own graphics card and decides what they do and think. Its PUBG Ally is a squadmate you talk to, and it had its first public test this summer.

The next ten years of games are going to be wild, the characters most of all. Someone in another thread said every character in a game could soon be run by AI, different every time and remembering everything you’ve done to them, with the story going wherever that takes it. Same core story, but everything around it different for every player. I agree.

Everyone’s a director now. The ones who’ll enjoy it are the ones who always cared more about the result than the code.

If you’d like to see what a week of direction turned into, Trazo is free to play in the browser, on a phone too. I’d love to know what you think of it, good or bad, through the contact form. And if you have an AI that fancies its chances, the board is waiting.