A ChatGPT Unity experiment that proves its own impracticality

What it does

AICommand is a Unity Editor window that pipes natural-language prompts to the ChatGPT API and attempts to run the resulting code inside the editor. It treats the engine like a voice-activated automation target, letting you describe changes in plain text instead of writing scripts by hand. The author built it specifically to see whether an LLM can reliably drive a complex creative tool.

The interesting bit

The project’s real output is not working software but a public post-mortem: the README openly declares the whole thing impractical and treats successes and failures as equal research data. That transparency is rarer than the integration itself, turning a fragile demo into a useful snapshot of where generative AI currently hits the wall.

Key highlights

• Generates and executes editor scripts from natural-language prompts via the ChatGPT API.
• The author explicitly labels it a proof-of-concept that “doesn’t work yet,” aimed at harvesting ideas from its own breakdowns.
• Targets Unity 2022.2 or later.
• Stores the OpenAI API key in UserSettings/AICommandSettings.asset, which risks accidental commits if ignored.
• Common failure modes include compilation errors from hallucinated code and NullReferenceException crashes tied to expired OpenAI billing.

Caveats

• Ships without a package or installer; the author notes it is only portable by manually transplanting the Assets/Editor folder.
• ChatGPT frequently writes broken code, and the documented recovery strategy is repeatedly pressing the Run button until the model guesses right.
• A NullReferenceException in AICommandWindow usually signals an expired OpenAI trial rather than a logic bug.

Verdict

Grab it if you are studying AI-assisted creative workflows or need a grounded reality check on LLM reliability. Avoid it if you are looking for a stable, production-grade Unity automation layer.