July 2026 Summaries

As AI agents increasingly become adept at retrieving information like professionals, the recent announcement by Perplexity of "search as code" marks a significant evolution in the field of information retrieval. Initially, the simple approach of using vector databases to chop text into chunks for retrieval proved insufficient, as it lacked context and relied solely on vector similarity. The subsequent hybrid search techniques, incorporating human information retrieval methods like BM25 and machine-learned ranking, improved search quality significantly. However, Perplexity's new approach suggests a third stage in search technology, allowing AI agents to execute complex queries that mimic expert search behavior, akin to a financial analyst or "quant." This approach circumvents the limitations of human users, who often provide vague search inputs, by empowering agents to conduct semantic searches, filter and rank results effectively, and leverage a broad array of search tools. This paradigm shift emphasizes the need for developers to focus on providing richer search capabilities for AI agents, rather than limiting them to the simpler techniques used for casual human searches.