What Is Agent-Based Modeling?
Agent-based modeling (ABM) is a computational method for studying systems composed of many interacting entities. Instead of writing equations for aggregate quantities like population size or infection rate, an ABM defines the rules that individual agents follow and lets the aggregate behavior emerge from their interactions. The output is not just a number — it is a full trajectory of every agent in the system over time.
This is a stub pillar page. The full guide will cover ABM history, when to use it, modern tools, and applications across epidemiology, social science, ecology, and transportation. The interactive browser simulators below let you explore the canonical models hands-on without writing any code.
Three Things ABM Does Well
- Heterogeneity. Every agent can have different properties — age, income, risk tolerance, location. Equation-based models struggle with this; ABM is native to it.
- Spatial structure. Interactions happen between neighbors, not a well-mixed pool. Epidemics move as wavefronts, fires propagate along gradients, markets form local clusters.
- Emergence. Trivial local rules produce surprising global outcomes. Boids flock. Schelling segregates. Sugarscape becomes unequal. Cooperation survives on lattices.
The Canonical Models
Most ABM introductions start with the same handful of models because each one isolates a single mechanism cleanly. All of them run in your browser below — drag the sliders and see the behavior emerge in real time.
Schelling segregation (1971)
Mild same-type preferences produce extreme spatial segregation. The first and still most famous demonstration that individual preference and aggregate outcome can diverge sharply.
Reynolds boids (1987)
Three rules — cohesion, separation, alignment — produce coherent flocking from purely local information. The template for all of swarm behavior modeling and VFX crowd simulation.
Conway's Game of Life (1970)
A cellular automaton with two rules that turns out to be Turing-complete. The simplest entry point to emergent computation and the universe of cellular automata.
Epstein-Axtell Sugarscape (1996)
Heterogeneous agents harvesting a regenerating resource produce Pareto wealth distributions — inequality emerging from individual-level rules with no policy or coercion.
When to Reach for ABM
ABM is the right tool when heterogeneity, space, or networks matter — and an equation-based model would require so many compartments that you lose the mechanistic interpretability. If you can write the system as a simple ODE and the dynamics match reality, do that. If you need to track who interacts with whom, you want ABM.
Tools of the Trade
- NetLogo — the teaching standard. Huge model library, excellent docs, easy onboarding. Limited for large-scale production work.
- Mesa — the Python ABM framework. Widely used in academic research. Scales to ~100K agents on CPU.
- FLAME GPU 2 — the GPU framework. Handles 10M+ agents in seconds by compiling agent rules to CUDA kernels.
- Repast HPC — the MPI framework. Used when you need cluster-scale simulations with agent counts exceeding a single GPU.