search ai-systems code-generation ai-agents

Rethinking search as code generation

research.perplexity.ai

June 2, 2026

23 min read

🔥🔥🔥🔥🔥

47/100

Summary

Search is evolving from monolithic services to programmable primitives to enhance AI systems. As frontier models advance, they require access to fresh, accurate, and well-curated knowledge, making search a foundational component for drawing conclusions and performing actions in AI products.

Key Takeaways

Perplexity introduced a new search architecture called Search as Code (SaC), which allows AI agents to access and control search primitives directly within their harnesses.
Traditional search systems are becoming outdated as users demand more complex, end-to-end task completion from AI agents, requiring a shift from monolithic architectures to more flexible, programmable search components.
The new architecture enables AI models to exert fine-grained control over search processes, enhancing their ability to retrieve and process information effectively.
Perplexity's search stack currently serves thousands of queries per second and is continuously optimized through self-improvement loops to better meet user needs.

Read original article

View story →

Coverage & discussions

Hacker News · research.perplexity.ai

Rethinking search as code generation ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Mixed

Positives

The proposed approach could significantly enhance coding agents' efficiency by enabling a more structured search strategy, potentially reducing the number of iterations needed.
Integrating a multi-stage search pipeline may streamline the process, allowing coding agents to fan out searches and filter results more effectively.

Concerns

There are concerns about the model's ability to understand complex codebases, which may hinder its effectiveness in generating accurate search queries.
The potential for customer support issues arises from the generated code not always executing queries optimally, which could lead to user frustration.