Simulation Data Moat¶
A durable competitive advantage claim from mikhail-parakhin: customer simulation is easy to prompt and hard to make real. Without years of genuine user behavior traces to calibrate against, simulated customers only do what the prompter wanted them to do — giving a confirmation-bias engine, not a test environment.
shopify's moat around simgym is not the simulator code (other teams can build that). It is the merchant + buyer behavior history that lets the simulator's outputs be validated against actual population patterns. Startups without that history cannot replicate it — they can only prompt and hope.
"If you're just a startup I wouldn't do it… without the data, whatever you say in prompt, that's what the agents will be doing."
Broader relevance¶
This is a specific instance of the more general install-base-moat pattern: platforms with years of user-facing data can build AI products competitors cannot clone even if the model architecture is open.