AICodeKing demonstrates Antigravity's boost workflow using a procedural maze game built with Gemini 3.8 Flash. The explanation distinguishes the selected model from the deeper multi-agent workflow and cautions that plan allowances, quota and credit settings determine access and cost.
The maze provides observable checks for walls, key collection, edited paths and solver resets. Independent review finds subscription and movement-validation bugs, leading to fixes and regression tests. The creator emphasizes repeatable examples and direct browser checks rather than assuming an agent's success message proves correctness.
Watch on YouTube
