Perplexity is introducing a new approach it calls Search as Code, a framework meant to make search pipelines programmable by AI models. The company is positioning the idea as a way to rethink how search systems are built and executed, shifting more of the search workflow logic into code-like instructions generated by models.

The announcement reflects a broader trend in AI product development, where models are increasingly used not only to answer questions but also to assemble the steps needed to retrieve and organize information. In Perplexity’s framing, search should be less of a fixed interface and more of a programmable process that can be shaped dynamically by a model depending on the task.

By describing the system as Search as Code Generation, Perplexity signals that it wants models to play a larger role in determining how queries are handled, how sources are selected, and how results are assembled. That could make search more adaptable in contexts where a single query may require multiple retrieval steps or different search strategies.

The concept also suggests a move away from rigid, manually designed search pipelines. Instead of relying only on predefined workflows, Search as Code would allow models to generate or coordinate search logic on the fly. That approach could appeal to developers looking for more control over search behavior without having to build every step by hand.

Perplexity has not, based on the available material, provided detailed technical specifications or rollout plans. The core message is the company’s attempt to reframe search as something that can be expressed and executed programmatically by models rather than treated as a static product feature.

That framing fits with Perplexity’s broader emphasis on AI-assisted information retrieval. The company has long positioned itself around search experiences that blend natural language interaction with source-backed answers. Search as Code extends that philosophy by focusing on the underlying machinery that determines how information is gathered in the first place.

The idea may also resonate with enterprise and developer audiences, where search often needs to be customized for specific workflows, data sources, or quality requirements. If models can generate search pipelines directly, organizations may be able to adapt retrieval processes more quickly to changing use cases.

At the same time, the approach raises familiar questions about reliability, transparency, and control. When models generate the logic behind a search workflow, developers may want clearer visibility into what steps were taken and why certain sources were chosen. Those issues are likely to matter if the concept moves from framing to implementation.

For now, Search as Code appears to be as much a product and technical philosophy as a concrete feature launch. Still, it points to a direction in which search systems become increasingly modular, model-driven, and programmable, with AI doing more than finding information. It would also help decide how the search itself should work.