Frontier research

Research & Experiments

A field guide to the AI signals starting to change software work, agent systems, synthetic worlds, and leadership decisions.

Each piece shows what is proven, what is being tested, and what still might break.

Start here

Start with the live experiments

These pieces show where the thinking gets tested under real constraints.

A public lab notebook on gpt-image-2, synthetic suburbia, continuity, object tracking, and the failure modes hiding inside generated errands.

Autonomous research outputs from end-to-end experimentation pipelines, including methodology and generated papers.

Decision guides for when to train, how to curate datasets, and how to create release gates that block weak models.

A procedural 3D island with generated terrain, ocean motion, cinematic fly-throughs, and free-roam exploration.

Research tracks

Choose the question closest to the decision you are trying to make.

What happens when a generated image world is asked to preserve a fictional person, a house, a phone, errands, and object state across time?

What happens when multiple AI agents collaborate on shared tasks without explicit coordination rules?

How should teams measure AI-assisted software delivery beyond superficial speed claims?

How can synthetic representations help leaders reason about systems, teams, and operating constraints?

Where do language models appear to reason, where do they imitate structure, and how should builders test the difference?

Can agent groups form persistent norms, and what does that imply for organizational simulations?

What informal behaviors emerge around formal structures, and how can leaders make better decisions with that reality in mind?

Look for the question, evidence, uncertainty, and current status before deciding how much weight to put on a claim.