Agents have bounded rationality — limited information, limited cognitive capacity, and use simple, transparent heuristics. This makes the model highly explainable, unlike black-box neural networks.

The economy is a Complex Adaptive System that is never in equilibrium. It constantly evolves, crashes, and adapts. Models are designed to capture crises, bubbles, and endogenous instability — not just steady states.

In climate-economy models (Macrocosm), the environment is a highly structured Input-Output Network. If Agent A (a microchip factory) is hit by a climate shock, the shock propagates through specific network links to Agent B (a car manufacturer).

Agents' decision rules are calibrated using real-world microdata. The model is validated by checking whether emergent macro-behaviour reproduces real-world "stylized facts" — fat-tailed stock returns, Pareto wealth distributions, and other empirical regularities.

His team developed BEAST (Big and Efficient Agent-based Simulation using Tensors), mapping agent interactions onto GPU tensor operations — enabling simulation of 30,000 firms and 160,000 assets in a fraction of the time.