Rethinking Search as Code Generation

Existing AI systems operated as a monolithic structure in which a model throws a query and the search engine returns results according to a fixed pipeline. However, in the era of AI agents performing complex and variable tasks, this approach is hitting its limits.

Agents perform hundreds or thousands of search operations even within a single task, and must directly control how information is retrieved, processed, and aggregated. The existing linear Function Calling approach cannot achieve this kind of fine-grained control.

To solve this, Perplexity is introducing the Search as Code (SaC) architecture. SaC exposes the components of the search stack as atomic primitives in the form of an SDK, allowing agents to directly design and compose the search process through code.

Through this, models can go beyond simply consuming search results and directly intervene inside the search stack to execute search strategies optimized for the task at hand.