📊 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 announced a new method called Search as Code (SaC), allowing AI systems to compose custom search pipelines dynamically. While promising, the approach builds on existing ideas and faces questions about independent validation and model comparisons.
Perplexity has introduced Search as Code (SaC), a new framework that transforms how AI systems perform search operations. This approach allows AI agents to assemble custom retrieval pipelines dynamically using code, rather than relying on fixed search endpoints. The development aims to improve accuracy and control in complex, multi-step tasks, positioning SaC as a significant innovation in AI search architectures.
On June 1, 2026, Perplexity’s research team published a detailed explanation of SaC, emphasizing that traditional search models treat search as a static process, which limits flexibility for AI agents executing multi-hour, multi-query tasks. SaC exposes core search functions—retrieval, ranking, filtering, and rendering—as modular primitives accessible via a Python SDK, enabling models to generate and run custom pipelines in a secure sandbox. This design allows for more precise control over search processes and the ability to implement complex logic, such as regex filtering or parallel fetches, on the fly.
The company showcased a case study involving vulnerability management, where SaC achieved 100% accuracy in identifying and characterizing over 200 CVEs, while reducing token usage by 85%, compared to traditional systems. Benchmark results indicate SaC outperforms existing models on four out of five tests, including the WANDR benchmark, where it achieved a 2.5× improvement over competitors. These results suggest SaC can deliver both higher accuracy and efficiency, especially in complex retrieval scenarios.
However, some skepticism remains. Critics note that the key benchmarks where SaC excels are either internally developed or not independently verified, raising questions about the generalizability of results. Additionally, the comparison involves different models and architectures, complicating direct evaluation. The core idea of turning search into executable code is not new; prior research and industry efforts, such as the CodeAct framework and Anthropic’s MCP, have demonstrated similar principles, indicating SaC is an evolution rather than a revolution.
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
Python SDK for search pipelines
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.
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.
Implications for AI Search and Agent Control
SaC represents a meaningful shift in how AI systems can manage search processes, moving from static endpoints to dynamic, programmable pipelines. This approach enhances control, enabling AI agents to execute complex, multi-step retrieval strategies that adapt in real time. If validated independently, SaC could lead to more accurate, efficient, and flexible AI applications across domains requiring high-precision information retrieval, such as cybersecurity, research, and enterprise search.
However, the reliance on proprietary benchmarks and the novelty of the approach mean that broader adoption depends on replication and validation. The potential to re-architect search stacks into composable, code-driven modules could influence future AI system design, but it remains to be seen whether other organizations can implement similar systems at scale.
Pro Tools Perpetual License NEW 1-year software download with updates + support for a year
Full version, permanent License of Avid Pro Tools. Includes 1-Year of software updates and upgrades.
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 turning search into executable code has been explored in academic and industry settings over the past two years. Notably, the CodeAct framework (ICML 2024) demonstrated that models trained on code perform better at orchestrating retrieval pipelines. In late 2025, Anthropic published MCP, emphasizing the importance of sandboxed code execution for scalable agents. Perplexity’s SaC builds on these ideas, re-architecting its search stack into atomic primitives to enable more flexible, programmable search pipelines. This development signals a broader trend toward integrating code execution within AI systems for enhanced control and accuracy.
While Perplexity claims a significant engineering achievement, the core idea is not entirely new, and previous efforts have shown similar benefits. The key innovation lies in the detailed re-architecture of search functions into composable primitives, which could be difficult for others to replicate without similar engineering resources.
“SaC offers a promising new way to think about search, enabling AI agents to craft custom retrieval pipelines dynamically, which could dramatically improve complex task execution.”
— Thorsten Meyer, AI researcher
API Phosphate Test Kit Box
Phosphate Test Kit Box
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Validation and Generalizability of SaC Results
It is not yet clear whether SaC’s reported performance improvements will hold up in independent testing or across different models and tasks. The benchmarks where SaC excels are either internally developed or not publicly verified, raising questions about their broader applicability. Additionally, the comparison involves different models (GPT-5.5 and Opus 4.7), making it difficult to isolate the impact of SaC itself versus underlying model capabilities. Further independent studies are needed to confirm these findings and assess scalability.
Independent Testing and Broader Adoption of SaC
Researchers and industry observers will likely focus on replicating Perplexity’s benchmarks and testing SaC in diverse, real-world scenarios. Open-source implementations or third-party evaluations could help validate claims and determine whether SaC’s approach can be adopted at scale. Meanwhile, other organizations may explore similar programmable search architectures, influenced by SaC’s conceptual framework, to enhance their AI systems’ control and accuracy.
Key Questions
What is Search as Code (SaC)?
SaC is a framework that allows AI systems to assemble custom search pipelines dynamically using code, rather than relying on fixed search endpoints, enhancing control and flexibility.
How does SaC improve search accuracy and efficiency?
By enabling models to generate and execute tailored retrieval pipelines, SaC reduces redundant queries, improves filtering, and adapts to complex tasks, leading to higher accuracy and lower token usage.
Is SaC a completely new idea?
No, the concept of turning search into executable code has been explored previously in academic research and industry projects, but Perplexity’s implementation is a significant engineering re-architecture.
Will SaC work with other models besides GPT-5.5?
It is unclear; current demonstrations involve GPT-5.5 and Opus 4.7, but broader compatibility depends on future development and integration efforts.
What are the main challenges facing SaC’s adoption?
Independent validation of results, scalability, and replicability of the engineering effort are key hurdles before SaC can be widely adopted.
Source: ThorstenMeyerAI.com
