Labline

• 

  Cover

Inspiration

I've been obsessed with AI for a while and the more you read about what labs actually deal with the messier it gets. Hosting models. Serving costs. Chasing benchmarks. Dealing with angry customer reviews. It's way harder than people think. At the same time I love tycoon games like Software Inc and Capitalism Lab where the economy feels real and every decision has a tradeoff. I wanted a game that puts you in the shoes of running an AI lab. A detailed economy simulator under the hood but still fun to play and educational enough that you walk away understanding why this industry works the way it does.

What it does

Labline is a strategy game in the spirit of Capitalism Lab but about the AI industry. You run your own lab and compete against five rivals to build the best models judged on price speed and benchmarks. Everyone shares the same economy so you're fighting over the same compute the same money and the same customers. You train models ship them through subscriptions or APIs keep your serving infrastructure alive and reinvest in research or physical stuff like data centers. Every shortcut you take tends to come back and bite you later.

How we built it

I started with a concrete plan. I knew from the getgo that I wanted a detailed shared economy so before any UI existed I had it research how other sim and tycoon games handle their economies and come up with formulas. We designed a shared system that both the player and the rivals play by. Same markets same rules. We built the economy simulator first and refined it until it was usable.

Once the sim was solid we worked in phases across the layers. Map then UI then stitching everything together. I had it spin up multiple sub agents each owning a small piece of the game like the initial UI and the map. That took around 2 days and burned a full reset. After that I spent time refining the UI using image gen and letting it cook pushing it to feel more like a game and less like a spreadsheet with buttons.

It's still really rough since I didn't have much time but it's a great starting point and I fully plan to keep building on it.

Challenges we ran into

Stitching broke a lot of features. Getting independantly built pieces to fit together was the hardest part of the whole project. On top of that Codex crashed a couple of times and I lost all my annotations and stacking too many annotations at once led to a lot of lost threads. That sucked not going to lie.

The turning point was learning to tell Codex to delegate. Split my mind dumps into separate threads and worktrees keep changes isolated review them one by one then merge. Once I got the hang of that workflow things got much better and way less chaotic.

Accomplishments that we're proud of

I've never had the balls to build a game from scratch even though I always wanted to. This was an amazing experience and I actually shipped something playable. Building the tooling behind the game like the economy simulator and the balancing was genuinely tough but Codex made it way easier to iterate on. The fact that the player and five rivals all run on the same economic rules and the game stays coherent is the part I'm most proud of.

What we learned

The biggest lesson is that Codex can turn one mind dump into many threads of work. I was able to tell it all my issues and ideas at once and it smartly split up the related work spun up worktrees made the changes there in isolation asked me to review each change then handled merging everything back into the main thread. Really impressive and it genuinely changed how I use AI.

I don't think of it as 1 to 5 parallel threads of unrelated work anymore. Instead I dump a whole week's or month's worth of ideas let it explore all of them at once then come back and review. Basically tinder swiping my ideas. It's a completely different way to build.

What's next for Labline

• Fix bugs and polish the UI/UX since it's still rough around the edges
• Add multiplayer and more rivals
• Bundle it into a proper desktop app
• Add more depth to training and make the simulation less deterministic so runs feel more alive

Built With

• bun
• next.js
• react
• three.js