About This Project
Overview
Scenario Lab turns a single prompt into a populated simulation: it spawns multiple subagents, runs them through a defined scenario, and reports back what happens to each one. Built with LangChain and PyTorch, it's designed for exploring second-order effects and surfacing AI-generated insight on situations that are hard to reason about from a single perspective.
Motivation
I wanted to explore whether spinning up multiple subagents to simulate a scenario could surface non-obvious second-order effects that a single-pass LLM prompt would miss.
Lessons Learned
Agent-to-agent interactions are unpredictable in interesting ways — small prompt changes to one subagent could cascade into very different outcomes for the others, which made evaluation harder than building the simulation itself.