How Anul Agarwal Vibe-Codes Web Games with Claude Code and Codex
Built by anul agarwal · · Source: X
Result: According to anul agarwal, their first 100 or so prototypes barely made money, while the next 20 or so got much closer to six figures in revenue. They say their games have now reached tens of millions of players, including 65M+ for a Watermelon/Suika variation.
What they built
anul agarwal says they have made games for about six years and built more than 120 prototypes for mobile, web and other platforms. Most are small 2D puzzle, casual, arcade or high-score games, and lately they have started making 3D games too. Examples they name include a Watermelon/Suika variation, Draw To Smash, an Element Merge variation, Cozy Rope Puzzle and Drop Animals.
How it works
They say AI now does a large share of the coding. They vibe-code games in Unity or HTML/JavaScript using AgentsGuild, Claude Code and Codex. Their workflow is:
- Scout ideas: They play popular games on app stores and web portals to learn the core mechanic and what keeps people playing. Then they change the theme or mechanics instead of making a copy.
- Prototype with AI: They give the agent references and explain the mechanics. They test what it produces and keep correcting it until the game feels right.
- Export per platform: According to the post, AgentsGuild exports builds with commands like /youtube, /poki and /crazygames.
- Pitch portals: They host a playable version on their own site, anul.space, and send the link to portals. Then they work through each platform's change requests and QA.
- Measure and update: They add their own analytics to see which level players quit on and whether they use rewarded ads. They also add new levels or seasonal content after launch.
They say they plan monetization while designing the game, mainly through rewarded ads. Longer play sessions make more money and also give platforms a reason to promote the game.
The result
According to anul agarwal, their first 100 prototypes barely made money and the next 20 or so got much closer to six figures. Their games have now reached tens of millions of players. They give the credit to distribution and to relationships with platforms, not to AI.
How to build something similar
- Choose a small, simple mechanic you can prototype quickly, and give your coding agent a reference game plus a clear description of the rules.
- Test each build yourself and send the agent specific fixes, such as controls, mobile layout or difficulty, instead of asking it to "make it better."
- Read each target portal's submission requirements before you build, so mobile support and ad placement are planned from the start.
- Add simple event tracking, such as level reached or quit point, so you base changes on data rather than guesses.
The original post
Vibe coding games but making $0? This article might help you! https://x.com/i/article/2106005420829716481 I have been making games for around 6 years now. Across mobile, web and other platforms, I have probably built 120+ prototypes. The funny part? The first 100 barely made me any money. The next 20 or so got much closer to 6 figures in revenue. Some examples: Watermelon/Suika variation — 65M+ players Draw To Smash Element Merge variation Cozy Rope Puzzle Drop Animals …and many more My ability to make games did not suddenly become 10× better. What changed was that I finally understood distribution. Instead of publishing a game somewhere and hoping people magically found it, I started working with web-game platforms that already had millions of players. Platforms like CrazyGames get millions of players every month Think CrazyGames, Coolmath Games, GameDistribution, Poki, YouTube Playables and similar portals. They already have the audience. Millions of players visit these websites every month. Your job is to build something their audience wants to play. Building the game is the easy part now Coming up with game ideas does not always come naturally to me. Usually, I start by scouting mobile stores and web-game portals to see what people are already playing. I look at games with lots of ratings, play a few of them and try to understand: What is the core mechanic? Why does it feel satisfying? How quickly can you understand the game? What keeps you playing? Where could the idea be changed or improved? The goal is not to clone the game. I usually take the core idea, change the theme, add or remove mechanics and put my own spin on it. Most of my games are currently 2D puzzle, casual, arcade or high-score games because they are small enough to build and test quickly. Although lately, I have also been moving more into 3D games. Racing and shooting games can work well on web too. For development, I use AI heavily. I vibe-code games in Unity or HTML/JavaScript using tools like AgentsGuild, Claude Code and Codex. AgentsGuild allows me to export builds for various web-platforms with simple commands like: /youtube, /poki, /crazygames etc. My workflow is basically: idea → prototype → test → feedback → iterate → ship AI now handles a huge amount of the coding. I give it references, explain mechanics, test what it produces and keep correcting it until the game feels right. My own games website! You can see some of the games I build on anul.space. For web specifically, engines like Phaser, Three.js, PlayCanvas or Cocos can be easier to work with. I still use Unity because I know it well, although Unity web builds usually require more optimization. Distribution is the actual business This is the part I misunderstood for years. Making a good game does not automatically mean people will play it. You need distribution. Web platforms solve that problem because they already have millions of users actively looking for games. Once I have a playable prototype, I start approaching platforms. Most portals have a developer or game-submission page where you can send them your game. Usually, I host a playable version on my games website and submit a link. The platform then evaluates things like: Is the game fun? Does it fit their audience? Does it feel different enough? Will people play it for long enough? Does it work properly on desktop and mobile? Can advertising fit naturally into the experience? Platforms have also become much more selective. Sometimes you submit a game and hear nothing. Sometimes they reject it. Sometimes they send you a long list of changes. That is normal. If I do not hear back, I may submit again later or contact someone from the company through LinkedIn or X. Once a platform is interested, there is usually an iteration process. They might ask you to change the onboarding, improve the controls, fix mobile responsiveness, adjust difficulty or rethink where ads appear. After that comes QA. You fix bugs, meet their t
Want to build something like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week.
Join Shipping Skool — Free