Perplexity AI unveiled Brain, a self-improving memory system for its Computer agent. Instead of only storing user preferences, Brain learns how tasks are executed — tracking what works, what fails, and how to improve outputs over time. It builds a context graph of all agent activity and continuously refines it, reviewing past sessions (often overnight) and updating its internal understanding. Early metrics: +25% task correctness on repeated tasks, +16% recall, −13% cost on context-heavy tasks. Rolling out to Max and Enterprise Max subscribers (Research Preview).