📊 Full opportunity report: Search as Code: Perplexity Is Right About the Future — Just Not First to It on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Perplexity has announced a new approach called Search as Code (SaC), allowing AI systems to build custom retrieval pipelines dynamically. This development aims to improve accuracy and efficiency in complex AI tasks, though some claims remain unverified and the idea is not entirely novel.
Perplexity has announced the release of Search as Code (SaC), a new architecture that transforms how AI systems perform search by enabling models to assemble retrieval pipelines in real-time. This innovation aims to address limitations in traditional search methods, particularly for complex, multi-step AI tasks, and represents a significant shift in search technology.
On June 1, 2026, Perplexity’s research team published a detailed proposal for SaC, arguing that conventional search systems are ill-suited for the agent era, where AI models need to execute multiple, dynamic retrieval operations. SaC exposes the core components of the search process—retrieval, filtering, ranking, and assembly—as atomic, programmable primitives accessible via a Python SDK. This allows models to generate and execute code that customizes the search pipeline for each task, rather than relying on fixed endpoints.
Perplexity demonstrated SaC’s effectiveness through a case study involving the identification and characterization of over 200 high-severity CVEs. Their system achieved 100% accuracy while reducing token usage by 85%, outperforming traditional systems that scored below 25%. The approach involves a three-stage process: fan-out over vendor advisories, refinement via an LLM, and verification, enabling bespoke, multi-stage retrieval programs rather than multiple endpoint calls.
Benchmark results show SaC leading on four of five tests, including WANDR, where it outperformed competitors by 2.5 times. The system also performs well at lower cost, with some configurations surpassing rivals in efficiency. However, some claims, such as the WANDR benchmark results, are based on proprietary datasets not yet independently verified, raising questions about their generalizability.
Search as Code
Perplexity says agents shouldn’t call a search engine — they should program one, composing atomic primitives into a bespoke pipeline in a sandbox. The thesis is right. It’s also the search-shaped version of an idea the field has been converging on since 2024.
Monolithic search
50 AI Agents Every Developer Must Build: The Complete Guide to Building Scalable, Production-Ready Autonomous Systems with LangChain, LangGraph, and Python
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Programmable primitives
Directionally right, genuinely engineered — the rebuilt-from-atoms search stack is the part rivals can’t cheaply copy. But it’s a strong execution of an industry-wide idea, validated mostly on benchmarks Perplexity ran itself. The moat is the infrastructure and the tuning loops, not the architecture.
Python SDK for search automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for AI Search and Retrieval Strategies
This development indicates a potential paradigm shift in AI search architectures, emphasizing programmable, modular retrieval pipelines that can be tailored dynamically. It could significantly improve AI performance on complex, multi-step tasks, reduce costs, and enable more precise control over retrieval processes. However, the approach’s novelty is contested, as similar ideas have emerged in recent research and industry efforts, suggesting that SaC is an evolution rather than a revolution.
For practitioners and organizations, adopting such architectures could mean more flexible and powerful AI systems capable of handling intricate information retrieval tasks more effectively. Nonetheless, the full impact depends on further validation, broader adoption, and how well the engineering principles translate into scalable, robust solutions.
RAG-Driven Generative AI: Build custom retrieval augmented generation pipelines with LlamaIndex, Deep Lake, and Pinecone
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of Search Architectures in AI
The concept of treating search as programmable code is not entirely new. Similar ideas appeared in the CodeAct paper (ICML 2024), which demonstrated that AI models perform better when they generate executable code to interact with tools, rather than relying solely on predefined APIs or tool calls. In late 2025, Anthropic published work on code execution with MCP, emphasizing the benefits of turning tools into code APIs within a sandbox environment. These efforts highlight a broader industry trend toward more flexible, code-driven approaches to AI tool use.
Perplexity’s innovation lies in re-architecting its own search stack into atomic primitives, a significant engineering effort that enables more granular control and customization. While the underlying idea is not entirely new, the implementation demonstrates a practical, scalable approach to integrating code-based retrieval pipelines into AI systems, which could influence future research and product development.
“SaC represents a meaningful step toward more flexible, controllable AI retrieval systems, though the core idea is rooted in ongoing industry trends.”
— Thorsten Meyer, AI researcher
Protecting User Privacy in Web Search Utilization (Advances in Information Security, Privacy, and Ethics (Aispe) Book Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unverified Claims and Benchmark Limitations
Some of the benchmark results, particularly WANDR, are based on proprietary datasets and have not yet been independently replicated. The comparison between different models running on different architectures also introduces variables that complicate direct evaluation. It remains unclear how well SaC will perform across a broader range of real-world tasks and datasets.
Further independent testing and validation are needed to confirm the scalability and robustness of the approach, especially outside controlled experimental settings.
Next Steps for Adoption and Validation
Expect ongoing efforts to replicate and validate SaC’s benchmark results by third parties. Industry observers will watch for broader adoption of programmable search pipelines in commercial AI systems. Perplexity is likely to continue refining its architecture, potentially releasing open-source tools or APIs to facilitate broader experimentation. Additionally, further research will explore how SaC compares to existing code-based approaches and whether it can be integrated into larger AI ecosystems.
Key Questions
How does Search as Code differ from traditional search methods?
SaC enables AI models to generate and execute custom retrieval pipelines in code, rather than relying on fixed search endpoints, allowing more flexible and task-specific search strategies.
Is SaC a completely new idea?
No, the concept of turning tools into executable code for better AI control has been explored in recent research and industry projects. SaC’s innovation lies in its specific architecture and engineering implementation.
What are the main benefits of SaC?
SaC offers improved accuracy, reduced token usage, and greater control over retrieval processes, especially for complex, multi-step tasks requiring dynamic search pipelines.
Are the benchmark results reliable?
Some results are based on proprietary benchmarks and have not been independently verified, so caution is advised until further validation is available.
What impact could SaC have on AI applications?
If validated and widely adopted, SaC could enable more precise, efficient, and adaptable AI systems capable of handling complex information retrieval tasks at scale.
Source: ThorstenMeyerAI.com
