Agents as Search Engineers // Santoshkalyan Rayadhurgam

Why It Matters

Agentic search reshapes how enterprises retrieve information, delivering higher relevance for ambiguous queries while keeping latency and cost manageable, and it sets the technical foundation for future AI‑driven knowledge assistants.

Key Takeaways

• Traditional search assumes fully formed user intent, now outdated.
• Agentic search adds stateful reasoning and multi‑turn interaction.
• Hybrid lexical‑vector backends with transparent tools improve determinism.
• Intent‑conditioned query embeddings boost precision by ~35% with low latency.
• Future horizons include domain agents, ambient intelligence, and AGI understanding.

Summary

Santoshkalyan Rayadhurgam argues that the foundational assumption of classic retrieval—users supply fully formed intent—is collapsing, prompting a transition from deterministic, stateless pipelines to agentic, stateful search systems that reason across turns.

He contrasts three generations: lexical BM25 pipelines, vector‑based RAG models, and the emerging agentic architecture. The latter treats retrieval as a control loop, maintaining session memory, dynamically selecting strategies (lexical, semantic, graph), and orchestrating multiple back‑ends with fault tolerance. This shift addresses high‑ambiguity, underspecified queries that require entity extraction, temporal grounding, and intent classification.

A concrete example—“find that Python memory thing from last week”—illustrates how a static engine fails, while an agentic system parses entities, resolves time constraints, and infers the needed artifact type. Rayadhurgam reports a 35 % precision lift using intent‑conditioned query embeddings with only ~200 ms latency, and shows a cost hierarchy from cache (≈10 ms) to full reasoning (≈500 ms).

The roadmap points to near‑term multi‑turn clarification, mid‑term domain‑specific micro‑agents, and long‑term ambient intelligence where search merges with understanding or disappears under AGI. For enterprises, adopting transparent, composable tools over monolithic black‑box APIs will be crucial to maintain determinism, debugability, and cost efficiency.

Original Description

March 3rd, Computer History Museum CODING AGENTS CONFERENCE, come join us while there are still tickets left.

https://luma.com/codingagents

Thanks to @ProsusGroup for collaborating on the Agents in Production Virtual Conference 2025.

Abstract //

Search is still the front door of most digital products—and it’s brittle. Keyword heuristics and static ranking pipelines struggle with messy, ambiguous queries. Traditionally, fixing this meant years of hand-engineering and expensive labeling. Large language models change that equation: they let us deploy agents that act like search engineers—rewriting queries, disambiguating intent, and even judging relevance on the fly. In this talk, I’ll show how to put these agents to work in real production systems. We’ll look at simple but powerful patterns—query rewriting, hybrid retrieval, agent-based reranking—and what actually happens when you deploy them at scale. You’ll hear about the wins, the pitfalls, and the open questions. The goal: to leave you with a practical playbook for how agents can make search smarter, faster, and more adaptive—without turning your system into a black box.

Bio //

Senior Engineering Manager - AI/ML

Engineering leader - AI for Ads at Meta

Ex- Lyft Pink Subscriptions

Ex - Amazon CV/ML

A Prosus | MLOps Community Production