Perplexity's Aravind Srinivas Reframes Retrieval as 'Search as Codegen' — New 'Search as Code' Replaces Tool-Calls With Generated Python

Storyboard18·medium signal

Perplexity launched Search as Code for AI agents (announced June 3), swapping traditional search tool-calls for a model that writes and runs Python scripts to orchestrate retrieval — what Srinivas calls 'search as codegen.' The bet is that as agents increasingly act through code execution, search should be a programmable primitive rather than a single rigid API call. For agent builders, it points toward retrieval logic (filtering, ranking, multi-step lookups) living in generated code instead of fixed tool schemas.