I built an airship game twice in a week
Not long ago I posted a video of the first time I asked my computer to build me a game. An airship game, part pirate ship and part zeppelin, with a painterly look, rock spires poking out of the clouds, a little economy and a crew.
Then I played it for a bit, thought "yes, but what I actually want is…", and started again.
The numbers
Version 1, which I called Stratum, took three days: 169 commits and about 26,000 lines of Rust, plus an Elixir server for the world.
Version 2 took four days: 299 commits, about 140,000 lines of Rust, 300 unit tests and around 4,000 lines of design docs.
Claude Code wrote the code. I couldn't have written that much in a week, and probably not in a couple of years of evenings either. What I did was decide what went in, play it, and send it back when it wasn't right. That's the bit I've always enjoyed anyway.
For version 2, I started by writing down everything I wanted changed from the first. Rock pillars became proper mountain ranges you can see 20 km across. The home-made physics was swapped for a real physics engine. The crew went from little minifigures to pixelly sprites, using my brother's baked-horde technique (which made me disproportionately happy). And alongside the keyboard command wheel, there are now pixel touch screens around the ship that you walk up to and poke.
The detail
Once building something costs this little, you stop rationing ideas. If I thought "wouldn't it be good if…", it went in.
You walk around your ship. Your crew have names and faces you can tell apart. You stand at the chart table, lay in a course, and the helmsman flies her there.
The world is procedurally generated. The capital has districts, a tram avenue and a cable car. Towns hold a market every third day and a feast day every ninth, and the townsfolk panic when the guns go off.
Pirates hail you on the wireless, and you can defy, bluff, threaten, surrender or negotiate. They're not always very good pirates.
Things go wrong in a lot of detail, too. Wounds are tracked per body part, so a broken leg gives you a limp, a concussion muddles your orders, and a bleeder gets worse until someone sees to it. The surgeon works out who to treat first. If a shot reaches the magazine, the whole ship goes up.
Nobody handed me a spec for any of that. It was mostly me saying "ooh, and…".
Teenage me
I know AI is meant to be the death of creativity, but I think that's nonsense.
As a teenager I had loads of free time and nowhere near the skills to do anything with it. I'd have built a Command & Conquer: Tiberian Sun clone, probably several. The odd god game. Almost certainly a Lincolnshire version of Grand Theft Auto: Grand Theft Combine Harvester, where you nick a tractor, outrun the police down the A15 and hide in a field of sugar beet.
None of those got made, because getting from the idea to a playable game meant years of learning engines, maths and art. Most people never get there, and the ideas stay ideas.
That's changing quickly. You still have to know what you want, notice when it's wrong and care enough to fix it, but you don't need the years of learning first.
It's not just games, either. The same goes for the tool your business is missing, or the app your club keeps talking about. Games are just the most fun example.
If you've got a game (or anything else) knocking around in your head, have a go. It's nearly as easy to build your own very specific thing now as it is to play someone else's.
Right, I'm off to find out whether combine harvesters can be armed.