Rethinking Search as Code Generation
Summary
Perplexity introduces Search as Code (SaC), a programmable search architecture that lets AI agents assemble and execute retrieval pipelines via an Agentic Search SDK inside secure sandboxes. The article covers the three-tier design (models, sandboxes, SDK), code generation and execution, autoresearch improvements, and a CVE-based case study with strong accuracy and token-efficiency results, followed by benchmarks and future directions for a hybrid compute approach.