Perplexity is transforming search with "Search as Code," a new architecture addressing limitations of traditional search pipelines for AI agents. The company argues monolithic search systems can't handle the complex tasks modern AI systems perform, requiring thousands of retrieval operations that traditional architectures cannot support.

Search as Code exposes search components as primitives within an SDK, allowing models to assemble custom retrieval pipelines through code generation. This approach gives agents direct control over each search step - retrieval, ranking, filtering, and rendering - while providing access to intermediate states like candidate lists and ranking signals.

The new architecture establishes a cost-performance frontier for agentic search, enabling agents to optimize pipelines in-flight and consume only the most useful information. This represents a fundamental shift from search as a fixed system to search as a flexible, programmable resource for AI agents operating in complex information environments.