Engines and runtimes
Benchmarking
Comparable game slices are used to test how different engines and runtimes behave under AI-assisted implementation, verification and iteration.
AI game development, tested in practice.
By JTT Games is researching how AI can support the full game-development process without replacing engineering evidence, playtesting or judgement. This page is the canonical home for that work.
What is being tested
Agents and production workflow
In development
Research covers how specialist agents can coordinate implementation, QA, evidence, release work and recovery without treating model confidence as proof.
Creator experience
Research stage
The long-term goal is a human-facing creation workflow that can be approachable for a non-technical creator while preserving a real editable project underneath.
Assets, provenance and publishing
Public foundation live
The asset library now has a catalogue contract, free and premium lanes, licensing/provenance rules and a local browser image inspector. Packs themselves remain unpublished until release evidence exists.
Evidence before claims
A source file existing is not the same as a game working. The current workflow separates structural checks, real runtime execution, visual and interaction QA, device/platform evidence and human playtesting. Public conclusions should state which level of evidence actually exists.
What will live here
- Research — verified advances, experiments and practical findings.
- Benchmarks — comparable engine, tool and agent results with methodology and evidence.
- Registry — maintained AI game-development tools, open-source projects and integrations.
- Guides — durable tutorials derived from problems actually solved.
- Assets — free and premium game-ready resources with transparent provenance, licensing and technical evidence.
- Create — reserved for the future human-facing AI Game Development OS once it is genuinely usable.
Current status
The underlying AI game-development pipeline and benchmark programme are active research and development work. The asset-library foundation is now implemented on this branch, while research findings, benchmark conclusions and downloadable packs will be added only when backed by reproducible implementation, execution evidence or clearly labelled primary-source research. Unknowns remain unknown until they are tested.