To build a defensible AI company, go beyond scraped web data (what people say) or behavioral data (what people do). The real moat is in creating causal models by collecting data from randomized control trials and A/B tests to understand *why* people act, which allows you to shape future outcomes, not just predict them.
Unlike foundation models aiming for super-rational intelligence (coding, math), Simile's models are designed to capture human behavioral nuances. They are trained to make the same mistakes and exhibit the same biases as real people, focusing on the subjective, irrational side of human decision-making.
Customers like Starbucks don't just want a prediction that Frappuccino sales will fall. They want to know what actions to take to prevent that from happening. The true value of simulation AI is providing a causal model that allows businesses to test interventions and proactively shape their future, not just passively observe it.
A powerful hiring heuristic is to identify candidates who, looking back at their career, were consistently the primary reason for success in different environments. This signals extreme ownership and an ability to reinvent themselves, making them highly valuable assets for a scaling company.
Truly exceptional talent often combines two powerful traits that shouldn't coexist, such as being both deeply data-rigorous and wildly creative, or intensely paranoid short-term yet religiously optimistic long-term. This rare blend of contradictory skills is a signal of a world-class operator.
Contrary to the belief that enterprise sales cycles are always long, Simile signed Fortune 500 companies in under three months. When a customer's pain is sufficiently acute and a solution is presented, large corporations will drop everything and move at lightning speed, shattering typical procurement timelines.
The next frontier of AI monetization involves extremely high-value, compute-intensive services. A single simulation session might cost $10-20 million to run but prevent a half-billion-dollar mistake for an enterprise. In this world, customers will willingly pay $100 million for that single outcome.
When evaluating founders coming from academia, VCs should distinguish between those fixated on a niche problem and those driven by real-world impact. The latter are more likely to find a problem that generates revenue and successfully transition from a science project to a viable company.
To solve for memory and create personality in AI agents, Simile developed a "reflection" process. At intervals, like a "shower thought," the agent synthesizes disparate low-level memories (e.g., eating omelettes five times) into higher-level conclusions about itself (e.g., "I am invested in this research topic"), shaping its identity.