June 2, 2026
Search party or search parody?
Rethinking Search as Code Generation
Perplexity wants search to act like a DIY robot brain — commenters aren’t fully sold
TLDR: Perplexity says AI needs a new kind of search system built from small reusable parts so agents can handle bigger real-world tasks. Commenters were intrigued but skeptical, with the main debate boiling down to: is this a real breakthrough, or just a fancier way to describe better search tools?
Perplexity just dropped a big idea: instead of search being one giant black box that spits out answers, it wants search to become a set of smaller building blocks that AI agents can mix, match, and run like little scripts. In plain English, the company is saying tomorrow’s AI won’t just ask one question and move on — it’ll do entire jobs, making hundreds or even thousands of searches in the background. Ambitious? Absolutely. But in the comments, the real show began.
The loudest reaction was a giant, blinking “wait, but how?” One commenter loved the concept but immediately poked at the weak spot: if an AI is supposed to build a fancy multi-step search plan, how does it know enough to make the plan in the first place? That skepticism became the mood of the room. Another person cut straight to the practical jab: is this really different from just offering a better search API? Ouch. That’s the kind of deceptively simple question that can haunt a launch post.
Then came the classic internet side quests. One commenter compared the idea to an existing code-search tool, basically saying, “Cool pitch, but doesn’t something like this already exist?” And in a beautifully chaotic moment, someone ignored the AI philosophy debate entirely to ask what software was used to make the flowchart graphics. Peak comment-section energy: half future-of-computing showdown, half design-tool detective hunt. The vibe overall? Curious, skeptical, mildly amused — and very ready to make Perplexity prove this isn’t just search, but with extra steps.
Key Points
- •Perplexity argues that search is a core infrastructure primitive for AI systems because it provides access to fresh and curated external knowledge.
- •The article says traditional search pipelines treat search as a fixed monolithic service, which was sufficient for earlier, simpler AI use cases.
- •Perplexity states that modern agents now perform long-running, highly variable tasks that can require hundreds or thousands of retrieval operations.
- •The company introduces Search as Code (SaC) as a reference architecture that exposes search building blocks as programmable primitives inside agent harnesses.
- •Perplexity says its search stack serves thousands of queries per second and has supported products such as Search API, Agent API, and Computer.