The Best AI Companies Have Unique Data Acquisition Strategies | Simile Co-founder & CEO

By 20VC

Categories: VC, Startup

Summary

AI companies need defensible data strategies to survive—Simile's founder reveals how LLMs trained on massive behavioral data can power realistic agent simulations, with potential $100M simulation sessions in 2-3 years. The key: extract human behaviors through memory, planning, and reflection architectures.

Key Takeaways

  1. LLMs contain vast human behavioral and sentiment data from web training—founders can extract realistic agent behaviors by 'poking at them at the right angle,' enabling domain-agnostic, generalizable human-like agents.
  2. Solve agent memory problems by implementing markdown text storage plus 'reflection intervals'—agents explicitly synthesize memory clusters (e.g., 'Why did I get omelette 5 times?') rather than treating each interaction separately.
  3. Multi-agent systems require explicit memory and planning architectures—early GPT-3.5 agents paired with memory/planning/reflection prevented agents from greeting the same roommate as a stranger repeatedly.
  4. Data defensibility is the fundamental moat for AI companies—unique data acquisition strategies determine which AI founders build sustainable, defensible businesses versus commoditized solutions.
  5. Self-organizing agent behaviors emerge at scale—25 NPCs in a game town autonomously planned Valentine's parties and decorated cafes without explicit programming, proving agent sophistication at small scale.

Related topics

Transcript Excerpt

I think there's a world in which in about 2 to three years we're running a single simulation session that people will pay $100 million for it. This is Jun Park, founder and CEO at Similey. They predict the future. They're a simulation market that try and predict future human behavior. It is incredible. >> My fundamental thesis here is for AI companies of this generation, you need to have an interesting data strategy that's going to be defensible. This was one of the best AI technical conversations we've had in a long time and it was incredible to have June on the show. People live through different stages in their life and they have different careers, different jobs. At each stage of their life, were they the reason why that thing was successful? If you squint, were they the common denomin…

More from 20VC