"AI writes the code, but a human has to care"

• AI tools have dramatically lowered the barrier to entry, enabling a non-programmer to launch a multiplayer game in under seven months.
• Simple, well-understood game mechanics are ideal for AI-assisted development, allowing rapid prototyping while leaving depth to emerge from player interaction.
• AI excels at building systems, but human judgement remains essential for “game feel”, product decisions, and understanding what players actually want.

This article was originally published in the AI Gamechangers newsletter, curated by Steel Media GM Dave Bradley.

Meet Jonathan Bjerk, a 38-year-old certified accountant from Norway who built Curve Clash, a real-time multiplayer party game for web and mobile, using nothing but AI tools and determination.

Inspired by the 1990s Amiga classic Achtung, die Kurve!, Jonathan went from zero programming experience to a published cross-platform title in under seven months, using ChatGPT, Cursor and Claude.

In this interview, he shares what “vibe coding” looks like day-to-day, why simple game mechanics suit AI-assisted development and what surprised him most about the unglamorous realities of shipping a game.

You can learn more about AI in games at Pocket Gamer Connects. The conference series is heading to Malmö on May 27th to 28th and Barcelona on June 15th to 16th.

AI Gamechangers: Can you briefly describe your own background with games and AI? You’re an accountant by trade, so how did you get involved in the mobile games scene?

Jonathan Bjerk: I’m a certified accountant and I still work full-time. I’ve always liked playing games, though, and grew up playing Achtung, die Kurve! on a single keyboard with friends, which is basically the game that inspired Curve Clash.

I had zero programming background beyond some self-taught HTML and CSS for simple websites. What changed everything was ChatGPT.

One night in May 2025, I started experimenting with it, just asking it what we could do together. I quickly realised that AI had reached a point where someone like me (someone with ideas and determination but no coding skills) could actually build something real. I started Curve Clash on 28th May, 2025 and never looked back.

Please briefly tell us what sort of game Curve Clash is. What genre does it fit in?

Curve Clash is a real-time multiplayer snake game: think Snake meets Battle Royale. You steer a growing trail, dodge other players’ trails, and try to be the last one alive.

It’s cross-platform across web, iOS, Android and macOS, supports 24 languages, and has modes ranging from casual solo play to a 64-player Arena tournament with elimination rounds.

The main inspiration is Achtung, die Kurve!, a classic PC game from the ’90s, where you’d crowd around one keyboard with friends. I wanted to bring that exact feeling to mobile and make it work online with players worldwide.

The Arena mode, with its changing challenges each round – speed, zigzag, shrinking map, thick trails – is my own twist to keep it competitive and unpredictable.

What did “vibe coding” actually look like for you in practice? How did you go from idea to something playable? What tools were involved, and what was your biggest challenge?

Day-to-day, it was a conversation. I’d describe what I wanted in plain language and the AI would write the code. Early on, I used ChatGPT to get the basics working, the trail movement, collision detection and the core game loop.

Then I moved to Cursor, which was a game-changer because it could see my entire codebase and make changes across multiple files. Now I work almost exclusively with Claude, which I find best at understanding complex systems and making architectural decisions.

The whole stack is Flutter and Firebase, which the AI suggested early on for cross-platform support. My biggest challenge was a rough patch in September and October, where I was stuck on multiplayer synchronisation bugs.

Real-time multiplayer is inherently complex – you’re dealing with latency, state management, edge cases that are hard to reproduce. AI could write the code, but debugging intermittent multiplayer issues required a lot of patience and back-and-forth.

What took me a month to figure out in June, I can now solve in a couple of days, partly because my own understanding has grown alongside the tools, but most of all, the tools.

Curve Clash is inspired by a very simple, old-school concept. Why do you think that kind of design worked particularly well for an AI-assisted development process?

Simple core mechanics are key. The basic game loop – a trail that moves, turns and collides – is something AI can grasp and implement quickly. You get a playable prototype fast, which keeps you motivated. From there, you iterate.

But I think the real advantage is that the simplicity is deceptive. The rules are simple, but the emergent gameplay is deep – cutting off opponents, reading their movements, finding gaps. That depth comes from multiplayer interaction, not from complex code. So AI could handle the technical side while the game design complexity came naturally from putting players together.

If I’d tried to build a story-driven RPG or a physics-heavy 3D game, I think I’d still be stuck on chapter one.

Which parts of the process required the most human judgement?

Game feel. How fast should the snake turn? How thick should the trail be? How long should the gaps in your trail last? These are tiny parameters that completely change whether the game feels fun or frustrating and AI has no intuition for that. I spent hours just playtesting and tweaking numbers.

The other big one is product decisions: what to build next, what modes to include, how to structure the Arena tournament format. AI is an incredible builder, but it doesn’t know what your players want. That still requires you to play your own game obsessively, read every piece of feedback and trust your gut.

Multiplayer and cross-platform support are usually major technical hurdles. What surprised you most about building those systems with AI tools?

How far you can get before you hit a wall, and how hard that wall hits.

AI got me to a working multiplayer prototype shockingly fast. Cross-platform was almost free because Flutter handles that natively, and the AI knew to suggest it early on.

But real-time multiplayer synchronisation across devices with different network conditions – that’s where it got brutal. The AI could write technically correct networking code, but the edge cases in a live environment are endless.

Players on spotty mobile connections, race conditions, state desync. That September to October period, when I was stuck on bugs, taught me that AI is phenomenal at building systems, but you still need to deeply understand what you’ve built when things go wrong.

Curve Clash climbed to fourth in Norway’s Family Games chart. What do you think drove that traction?

We’re at over 1,500 players now and what I’m most proud of is the engagement – over 15,000 minutes of gameplay. That’s a high average per player, which tells me people aren’t just downloading it; they’re actually playing.

The #4 in Norway’s Family Games chart was during launch weekend, so I won’t pretend that lasted. But honestly, I think the traction comes from the gameplay being genuinely fun and instantly understandable. You see someone playing for three seconds and you get it. The AI story is a nice hook for press and fellow developers, but players don’t care how a game was made – they care if it’s fun.

And the local multiplayer mode, where you can play with many people using phones as controllers, has been amazing for word-of-mouth in schools and at events.

Have you iterated much since launch? Have AI tools affected your ability to respond to player feedback or fix issues quickly?

Massively. Since launching the beta in August and the full release in mid-December, I’ve added Arena mode, Weekly Challenges, an achievement system, league rankings and countless balance tweaks. The iteration speed is the real superpower of building with AI. A feature that would have taken me a month in the early days now takes a couple of days.

When players give feedback, I can often ship a fix or improvement the same week. That responsiveness builds trust with your community. They see their feedback matter; they stick around. AI hasn’t just helped me build the game – it’s helped me maintain and grow it at a pace that would normally require a team.

As someone coming from outside the games industry, what parts of game development turned out to be harder than you expected? And where did AI genuinely level the playing field for you?

Marketing. By far. Building the game was hard, but getting people to discover it is a whole different challenge. I’m currently running a DM outreach campaign to gaming creators on TikTok – I’ve reached out to over a thousand of them to promote our upcoming Creator Arena Tournament.

The other surprise was how much time goes into things that aren’t the game itself – App Store listings, screenshots, localisation, server infrastructure, analytics, compliance. The unglamorous stuff.

Where AI levelled the playing field is clear: the actual coding. I went from someone who couldn’t write a line of Dart to having a published cross-platform multiplayer game in seven months. That simply wasn’t possible two years ago.

There’s some scepticism in the industry around “AI-made games”. Talk us through your philosophy about originality and human creativity while working with AI-generated code and systems. Were you mindful that some players are pushing back against AI?

I think the scepticism is understandable and partly justified. If someone uses AI to churn out a low-effort clone and dump it on the App Store, that’s not good for anyone. But that’s a problem with intent, not with the tool.

My philosophy is simple: AI writes the code, but a human has to care. I designed every game mode, tuned every parameter, playtested obsessively, and made every product decision. The creativity – the vision for what Curve Clash should feel like – is entirely human. AI was my engineering team, not my creative director.

As for players pushing back, I’m fairly open about using AI because I think it’s an honest and interesting story. But at the end of the day, players judge a game by how it plays. Nobody opens Curve Clash and thinks about how it was made. They think about whether they can beat their friends.

If another non-technical creator wanted to build a small multiplayer game today using AI tools, what would you actively warn them not to do, based on your own mistakes or dead ends? Please share some advice.

First: don’t start with multiplayer. Seriously. Get your core game loop feeling great in single-player first. I built Endless mode before touching online play, and that foundation saved me. If your game isn’t fun against AI, it won’t be fun against humans.

Second: don’t switch AI tools constantly. Each tool has a learning curve and you get better results the more context it has about your project. I went from ChatGPT to Cursor to Claude and each switch had a transition cost. Find what works and commit.

Third: scope ruthlessly. AI makes it tempting to add features because they’re suddenly “easy”. But every feature is something you have to maintain, debug and balance. My best decisions were things I chose not to build.

And finally: playtest with real people early and often. AI can’t tell you if your game is fun. Other humans can.

If you could make one prediction for the future of AI and games, what would it be? We’re interested in how you see vibe coding experiences like yours changing the industry in the long-term.

My prediction is that within two to three years, the solo developer will become a legitimate force in mobile gaming. Not hobbyists making simple apps, but individuals shipping polished, competitive multiplayer experiences that would have required a team of ten a few years ago.

The barrier to entry is collapsing. What used to require years of [computer science] education and a team of specialists can now be done by someone with a clear vision and the persistence to see it through. That’s going to mean more games, yes, but also more diverse games – made by people from backgrounds the industry has never heard from. Accountants from Norway, for instance.

The studios that adapt will thrive. The ones that dismiss AI-assisted development will be competing against an army of passionate solo creators who can move faster than any committee.