Grok 3 and AI-Assisted Game Prototypes: A Tour of Creator Demos

I showcase browser games and interactive art that their creators associated with AI-assisted coding around the February 2025 release of xAI's Grok 3. These are selected creator demonstrations, not a controlled comparison of models. Browserbase sponsored a separate segment of this video promoting Stagehand.

The first example is Pieter Levels' browser flight simulator. It shows a plane, weather and day-night changes, speed and altitude indicators, shooting and a shareable multiplayer link. I find it impressive, though the demonstration also shows the plane passing through another plane, so the physics and collision handling are not complete.

Levels said he built an initial simulator in about three hours with Cursor, iterating from instructions and an HTML file. I introduce it as made with Grok 3, but his cited post attributes the original build to Cursor and does not establish that Grok 3 alone produced the entire game. The tools and stages involved should not be collapsed into a single-model claim.

Next, a creator identified as Danny Limanseta shows Mech and Cheese, a survival-style game in which a robot shoots cheese-like enemies and gains levels. I also show a Plants vs. Zombies-style defense game, with turrets, walls, upgrades, solar power and incoming enemies. These examples illustrate the developers' reported workflows, not a guarantee that any prompt will yield a polished game.

Álvaro Cintas demonstrates a vertical jumping game with selectable sprites and difficulty levels. He describes creating a basic version first, then adding assets and refining the result. That incremental process is useful: ask for a small working game, inspect it, and add features rather than requiring every detail in one initial prompt.

A creator using the name Lord Asado presents interactive generative art: strands on a phone respond to touch, with controls for thread count, stiffness, gravity and force. The creator describes using Grok 3 both to help formulate the prompt and suggest controls. I see it as a creative use of the same iterative approach beyond conventional games.

Another example is Cyber Manhattan, a rotating cube containing a stylized three-dimensional city. I like the moving stars and visual effect, while noting that the scene is relatively simple and could be extended. The video presents it as a one-prompt example; its source and process are creator-reported.

A 3D maze example shows a character moving around a generated map. Its author supplied a detailed prompt covering libraries, scene setup, maze generation, walls and collision behavior. I suggest it could become a shooter or puzzle game, but those additions are ideas rather than features shown in the example.

I copy the supplied maze prompt into Grok with its thinking mode enabled. After roughly 137 seconds it produces code; I see errors in the editor but save and open the result anyway. The character moves and collisions work in my limited demonstration. I have not resolved or characterized every reported error, and a single successful run does not show reliable performance on every prompt.

Another creator presents a Wordle-style game with seven-letter words and seven attempts. The familiar color feedback distinguishes correct positions, letters in the wrong position and letters absent from the word. The creator reports generating it in about 90 seconds and shares the prompt and conversation; this is a reported example, not a timed independent benchmark.

Finally, I show a VR shooter with approaching red blocks and projectiles. The creator describes it as generated in one go, but I do not have a VR headset and cannot verify the shooting controls in my browser preview. I can move around the 3D scene; turning it into a complete, polished game would require further work.

The examples make me optimistic about AI-assisted game prototyping, and I would like to build another game in a future video. They also show that prompting, iteration, generated assets, coding tools and human review all contribute to a result; the demonstrations do not establish that every game can be built reliably without development work.