Vision
An engine designed around the AI's loop
Open Terrain Engine is being built so an agent can author a complete, shippable desktop game in idiomatic C# — and verify it the way a careful engineer would.
The problem
LLM agents can't build real games today — not for lack of intelligence, but because they're missing two senses. Editor-based engines hide the game behind a GUI the agent can't drive. Code-only engines let the agent write rendering code but leave it blind: it can't see the result or test whether the jump feels right.
The thesis
An AI-native engine is built around a closed author → introspect → verify loop. Everything is text and code, the world is introspectable as structured data, verification is deterministic and headless, and the engine ships its own machine-readable API surface so the agent always knows the full interface.
Determinism is the keystone
Fixed timestep plus seeded RNG makes a game a replayable, assertable simulation. The agent reads back exact state and a screenshot a multimodal model can look at — and that same property unlocks lockstep netcode and replays for free.
Why C# is the wedge
One unified, statically-typed representation with the compiler as a free verifier, mainstream training mass so models are genuinely fluent, and no editor that owns the truth. A developer inherits world-class .NET tooling and can refactor anything the agent wrote.
This describes the target engine. Open Terrain Engine is in active development — design complete, implementation under way. Deeper feature breakdowns will land here as the build progresses.