Module 4 of 6 · Intermediate · 10 min read

Game development

Prototype mechanics, write game code, generate assets and design levels with AI, and keep the fun under your control.

Key takeaways

1AI is excellent for fast prototypes: get a playable loop before polishing anything.
2Name your engine and version; engine APIs differ and change.
3Generated art and audio need a consistent style guide and a licence check.
4Playtesting decides what is fun. AI can't feel it for you.

01Where AI helps

The fastest wins come from getting something playable on day one.

Prototyping mechanics quickly in an engine or in the browser.
Writing and debugging gameplay scripts, shaders and tools.
Generating concept art, textures, icons, sound effects and placeholder music.
Drafting lore, dialogue, quest text and item descriptions.
Balancing: spreadsheets of stats, simulated outcomes and difficulty curves.

02The workflow

Resist generating art before the game is fun; it is easy to polish a prototype nobody enjoys.

Write a one-paragraph pitch and the core loop: what the player does every 30 seconds.
Prototype the core loop with placeholder shapes. Play it.
Iterate on feel (controls, timing, feedback) through short play sessions.
Add content and systems one at a time.
Create a style guide, then generate or commission assets to match it.
Playtest with others, collect notes, and use AI to triage them.

03Tools to reach for

State your engine and version in every prompt, because APIs change between releases.

Coding assistants and agents that know your engine's scripting language, such as C# for Unity, GDScript for Godot or C++ and Blueprints for Unreal.
Browser frameworks (canvas or WebGL libraries) for the quickest playable prototypes.
Image models for concept art, textures and UI icons; text-to-3D tools for rough props.
Audio models for sound effects, ambience and placeholder music.

04Prompt: a playable prototype

Ask for the smallest thing you can play and judge.

Example

Build a playable prototype of this core loop in a single HTML file using the canvas API, no libraries: {{core loop, e.g. the player dodges falling blocks and collects coins; speed increases every 10 seconds}}. Include keyboard and touch controls, a score, a game-over screen and restart. Put all tunable values (speed, spawn rate, gravity) in a CONFIG object at the top so I can tweak the feel.

05Prompt: engine scripting

Give the engine, version and the existing setup so code fits your project.

Example

Engine: {{engine and version}}. I have a player scene with {{nodes or components}}. Write a script for {{mechanic, e.g. a double jump with coyote time and jump buffering}}. Expose tuning values in the inspector, comment the non-obvious parts, and explain how to attach and test it. Use only APIs available in this engine version; if you are unsure whether an API exists in this version, say so.

06Prompt: content and balance

Use AI for volume, then edit for quality and consistency.

Example

Our game is {{pitch}}, tone: {{tone}}. Create 12 items as a table: name, rarity, effect, cost, and a one-line flavour text under 15 words. Keep power within these bounds: {{rules}}. Then check the table for items that are strictly better than another item at the same cost, and adjust them.

07Pitfalls

Fun comes from iteration on feel. Short play-tweak-play cycles matter more than any prompt.

Outdated APIs: models may suggest functions from older engine versions. Check the docs.
Inconsistent art: assets generated without a style guide clash. Fix the palette, perspective and line style first.
Licensing: check the terms of any tool you use for commercial assets, and the rules of the stores you plan to sell on.
Over-scoping: AI makes it tempting to add features. Finish a small game first.

Tip

Ask for tunable values in one config block. Most game feel comes from numbers you will adjust dozens of times.

Finished reading?

Mark it complete to track your progress, then test yourself with the Build with AI: Software & Media quiz.

Next: Automation and workflows →