Editor’s note: Nina Lopatina is speaking at ODSC AI West this October 28-30. Check out her talk, Evolving from Agentic RAG to Agentic Searchthere!

The way AI systems find and process information is undergoing a fundamental transformation. What started as simple retrieval-augmented generation (RAG) has evolved into something far more sophisticated: agentic search systems that can autonomously navigate complex information landscapes.

The Evolution Beyond Traditional RAG

Traditional RAG systems follow a straightforward pattern: parse+chunk+embed documents, store them in a vector database, retrieve similar chunks, and feed them to an LLM. While effective for many use cases, this approach is inherently limited. It’s static, relies on pre-indexed content, and struggles with nuanced information needs.

Agentic RAG introduced intelligence to the retrieval process by allowing the LLM to dynamically add additional retrieval steps as needed. Instead of passively fetching documents, these systems reformulate queries, chain multiple retrievals, and decide when additional context is needed. This approach has been increasingly adopted over the past year to improve the range of queries a RAG system can answer, and significantly improves response accuracy. But the evolution now is continuing to agentic search: systems that transcend document retrieval entirely.

Query Reformulation: Teaching Systems to Think

A key differentiator of agentic systems is query reformulation. If you ask about “What’s the ROI comparison between our Q3 marketing spend and the industry benchmarks for SaaS companies with similar ARR?” traditional systems search for those exact terms. But no single chunk would match this embedding: this query requires information across internal financial documents, industry research reports, and company metrics documents. Agentic RAG allows systems to cast a wider net, finding highly relevant information that might not match exact query terms. If this query is broken into separate subqueries, and this information is available in the system’s vector database, then agentic RAG can answer this query. But what if this information wasn’t all previously ingested and readily available?

Contextual you have’s query path demonstrates where query reformulation lies within the query path, the steps between an initial query from a user and the final retrievals

Context Layers: Multi-Channel Intelligence

The revolutionary aspect of agentic search comes from advancements in context engineering. Unlike traditional RAG’s limitation to pre-indexed documents, context engineering enables dynamic information gathering through the following tools:

• Web searches for real-time information
• API calls for structured data
• Direct file processing on-demand
• Live data stream integration

This multi-modal approach means systems aren’t confined to pre-processed content; they gather information dynamically from wherever it exists.

Real-World Impact

Consider a financial analysis agent. Traditional RAG searches static reports, often providing outdated information. An agentic search system takes a different approach entirely.

When analyzing tech stock performance, it reformulates the query to capture various aspects, then simultaneously pulls real-time market data, searches recent news for sentiment, accesses fresh regulatory filings, and monitors emerging trends. The system adapts its strategy based on findings – unusual volatility might trigger expanded searches into geopolitical news or sector analysis.

This adaptive approach delivers insights no single retrieval method could achieve alone. Combining RAG with agentic search is like doing deep research with your own data, with the flexibility to augment the context your LLM has access to on the fly.

Looking Forward

The evolution from agentic RAG to agentic search represents a paradigm shift in AI information interaction. By combining intelligent query processing, dynamic retrieval strategies, and multi-channel context gathering, these systems tackle increasingly complex information needs while maintaining efficiency.

Want to dive deeper? Join me at ODSC West on October 30 where I’ll demonstrate live implementations and share practical learnings from production deployments. We’ll explore how context layers support increasingly autonomous agents, essential knowledge for building next-generation AI applications.

About the Author

Nina Lopatina leads Developer Advocacy at Contextual you havea startup empowering developers to rapidly build accurate, scalable RAG agents and agentic search solutions that transform unstructured data into applications. She bridges product, content, and community to enable developer success, bringing seven years of hands-on experience as a developer and leader in NLP and language technology.