Rebuilding Reality with Agent-Based Models
Introduction
Agent-based modeling (ABM) has become an unexpectedly clarifying lens for how I think about complex economic and technological systems. It isn't just a tool for simulation—it's a philosophical stance. The core idea is this: you can't abstract away emergent behavior. If you want to truly understand the system, you have to rebuild its micro-foundations. This memo captures the shift in my perspective through recent modeling efforts.
ABM as a Better Fit for Complex Questions
I used to try breaking systems into high-level summary metrics: average market share, churn rates, adoption curves. But real-world systems don't obey linear approximations. Markets are shaped by bounded agents, asymmetries in information, lock-in effects, herd behavior, and policy inertia. ABM gives me a way to encode all that.
I keep coming back to one insight: emergent outcomes can't be faked. You can only "get" path-dependence, self-reinforcing feedback loops, or tipping points if your agents act in a way that reproduces them.
Case Studies from My Own Projects
Modeling AI App Uptake in Healthcare
Health system buyers behave differently depending on peer behavior, internal ROI thresholds, and contract lock-ins. These aren't bugs; they're features of the system.
When I introduced heuristics like "go with the herd" or "copy the most profitable peer," I started seeing much more realistic diffusion patterns.
Simulating Competitive Dynamics Among CROs
Traditional firm modeling assumes rational optimization and continuous strategy space. In reality, CROs evolve strategies via imitation, inertia, and partial information.
ABM let me build in reactive strategies: copy pricing of competitors, expand capacity only after seeing three cycles of unmet demand, etc.
Vendor-Buyer Feedback in EMR Markets
Adding contract length as a lever wasn't just cosmetic. It created second-order effects—stability for incumbents, fragility for startups, and oscillations in price pressure.
Why This Approach Changed How I Think
Building an ABM forces me to ask: how does the behavior I see emerge from actual decision rules? This has rewired how I approach strategic questions. Instead of "What is the optimal price?" I ask "What kind of pricing rules survive competition and customer response over time?"
This means spending a lot of time, like A LOT, trying to understand how each agent thinks about their decisions—including figuring out their heuristics.
It also disciplines my assumptions. When a model outcome surprises me, I can't wave it away—I have to trace back the agent logic, and either fix the rule or embrace the implication.
Looking Ahead
I see ABMs not as predictive machines but as structured environments for stress-testing beliefs. Every new simulation is an invitation to ask: "What if the agents acted differently? What invisible pressures shape their behavior? What feedback loops are we missing?"
In that sense, ABMs are not just models—they are arguments. And like any good argument, they are only as strong as their assumptions, their logic, and their ability to make us see familiar dynamics in new ways.