Until recently, even the simplest mobile game required knowledge of a programming language, the installation of development tools, and many evenings spent hunting for bugs. Now all you need is the ability to clearly describe what you want. Claude, a model from the American company Anthropic, can turn a simple description in a chat into a playable game that runs directly in a browser or mobile app. Journalists from several international publications tried it for themselves, and their experiences clearly show what works right away and where you may get stuck. The following guide will take you through the entire process, from the initial idea to a link you can send to friends.
What you will need
No special software or powerful computer is required. All you need is an account on Claude.ai and a web browser. After signing in, select Claude Opus 5.5 from the model menu. Then go to Settings and, under Capabilities, check whether Code execution and file creation is enabled. Without it, Claude would only describe the game rather than run it.
Journalist Eric Hal Schwartz from TechRadar noted after his experiment that he did not have to install anything, touch a single line of code, or leave the chat window even for a moment. That is exactly how it should work for you as well.
Start with the simplest possible idea
The biggest mistake beginners make is trying to create a large game with lots of levels on their first attempt. A small, clear idea works much better. You could try making a character jump by tapping the screen, dodging falling objects, a memory-matching game, an endless runner, or a reaction-speed test.
If you cannot think of anything, feel free to ask Claude itself for suggestions. Jeremy Caplan, who wrote about creating apps in Claude for Fast Company, also recommends drawing on your memories and remaking a game you loved as a child. He used this approach to create a word game and a simple breathing exercise app, all without any programming knowledge.
A prompt that works
Open a new chat and describe the game. The more specific your prompt is, the fewer corrections you will need to make later. Jose Antonio Lanz from Decrypt confirmed after his experiment that a detailed, well-structured prompt produces significantly better results. You can use the following template and simply fill in the title and main gameplay concept:
“Create a game for a mobile browser called [title]. The player’s task is to [what the player does]. Use large touch buttons, a vertical layout adapted to a phone screen, a score counter, a new game button, gradually increasing difficulty, simple animations, and clearly distinguish between winning and losing. The entire game must be playable inside an Artifact in Claude without any additional setup. Keep the code as simple as possible and optimize it for phone screens."
The prompt does not have to be in English, because Claude understands many languages, including German and Czech. However, explicitly ask it to create the result as an interactive Artifact rather than just a piece of code. An Artifact is a window next to the chat where the app runs immediately and can be tested right away.
The first version almost never works perfectly
Once Claude finishes the game, open it and play through an entire round. Start it, play it, and check whether the score is counted correctly. Let yourself lose or win, and then try starting again. Click every button, because that is where bugs most often hide.
Lanz experienced this right from the start. He wanted a game in which words fall down the screen and the player has to type them before they reach the bottom. In a few minutes, Claude created a project that looked like a finished game, but the start button did not respond. Even the first fix did not help. Lanz still did not touch the code and simply told Claude that nothing happened when he clicked the button and the game would not start. On the second attempt, everything worked.
This leads to the most valuable advice in the entire guide. You do not need to understand why something is not working. Simply describe in ordinary language what you see and what you expected to happen, then leave the debugging to Claude. Lanz communicated with the model this way a total of fourteen times before he was satisfied enough with the result to feel comfortable sharing it.
Test the game on your phone
Something that looks great on a monitor may behave completely differently on a phone. Open the same Artifact in the Claude app on your phone and play the game with your finger. Check whether the buttons are large enough to use comfortably, whether the entire game fits on the screen, whether the text is readable, and whether everything runs smoothly.
This step is worth doing. Lanz’s game worked for his colleagues in the newsroom, but only on a computer. No one could play it on a phone.
Improve it in small steps
When you find problems, return to the chat and describe them as specifically as possible. Instead of making a general complaint that the game feels strange, write, for example, that the restart button is too small on a phone and overlaps the score. Lanz advises against asking for everything at once and recommends building the game gradually, as if adding layers to a rough foundation. Fix one bug first, then another, and only after that add new features.
Once the game works reliably, add one round of improvements. Ask for an option to turn off the sound, save the high score, add a start screen, make the increase in difficulty smoother, and include a few simple visual effects. Then test the game on your phone again, because every change can break something.
Schwartz demonstrated what this kind of improvement looks like in practice. He asked Claude for a Super Mario-style game, except that instead of a plumber, the main character would be a journalist. In place of the mushroom that makes the character grow, he suggested a pen, and instead of a fire flower, a microphone that shoots flaming question marks. Within thirty seconds, he had a game called Press Quest. Its graphics did not come close to those of the original Mario, and the enemies moved strangely. Schwartz had also forgotten to describe how to advance to the next level, so his hero remained trapped in the first stage.
But a few sentences were enough. He asked for a newsroom building to serve as the level exit, adjusted the enemies’ movement and the placement of power-ups, and requested better graphics as well. It was done in a minute. When he discovered that all the levels looked the same, he had Claude add new ones, and within two minutes each one was different. Not everything worked out: some levels did not make much sense, and a piece of text remained outside the game window that he could not persuade Claude to remove. Still, the entire process from idea to playable version took him less than an hour.
Caplan describes a similar experience. Sometimes Claude leaves out a detail, a button does not work, or the result does not look exactly as intended. According to him, however, it handles requested fixes well.
Send the game to your friends
When you are happy with the game, click the Share button in the Artifact and create a link. Anyone who receives it can open and play the game without installing anything. According to Schwartz, other people can also customize someone else’s creation and make their own version of it.
Sharing can also bring unexpected fun. Lanz sent his word game to colleagues in the newsroom, and in addition to giving him ideas for improvements, they immediately took the high-score record from him. His colleague Ryan Gladwin scored more than 2,800 points, and the game’s creator now hopes to reclaim first place soon.
Sources: fastcompany.com and decrypt.co
