Most AI memory implementations focus on storage and retrieval. The inference—what to actually store and why—is superficial. This talk introduces a different approach: solve the inference problem first by training reasoning models to produce formal logic. It's the hardest reasoning for humans, but LLMs excel at it—even more so when trained. Build the storage and retrieval system around scaffolding that logic to produce comprehensive, evolving representations. Vince Trost (Co-Founder, CEO of Plastic Labs) walks through how to use Honcho, how it reasons over data, and how developers can leverage that reasoning to solve memory, build stateful agents, and focus on building the best AI products possible. It's simple to implement, come see how.