Sim-In-Silico simulates markets with AI agents

This is one of the more credible takes on synthetic populations: the value is not in asking an LLM whether it would buy, but in building a constrained world where memory, network effects, and time actually matter.

• –The three-layer design is the right abstraction: market context, persistent agent state, then LLM reasoning on top
• –The social graph matters as much as the model; without diffusion and clustered information, word-of-mouth simulations are mostly noise
• –The skeptical-cluster trial result is the kind of directional signal this category can produce, especially for pricing and onboarding hypotheses
• –The main risk is calibration: if agent behavior is not tied to real-world distributions, the output becomes a persuasive story rather than a decision tool
• –This looks most useful for early-stage GTM, pricing, and positioning tests, not for precise demand forecasting