Lab Notes

Field notes from the edge of AI and the physical world.

Essays on the systems, operating models, and domain judgment required to turn frontier AI into real-world capability.

Context graphs are the memory layer for agentic enterprise systems

AI agents do not fail only because the model is weak. They fail because the enterprise context is scattered across tools, conversations, decisions, and memory that the system cannot reason over.

World models are the training grounds for physical AI

Robots need their own version of the internet-scale data flywheel. World models, simulation, synthetic data, and in-robot experience are how that flywheel starts.

The domain-native AI organization

Why the advantage in vertical AI comes from operationalizing expert judgment, not just choosing a better model.