Agentic Search for Context Engineering — Leonie Monigatti, Elastic
TL;DR
Leonie Monigatti from Elastic argues that context engineering is fundamentally 80% agentic search, evolving from rigid RAG pipelines to dynamic agent-driven retrieval that must navigate diverse context sources through carefully curated, specialized search tools.
🔍 Evolution of Search Architecture 3 insights
From fixed pipelines to agentic decisions
Early RAG forced retrieval on every query regardless of necessity, while agentic systems let LLMs decide when retrieval is actually needed and support multi-hop reasoning.
The context engineering paradigm
Effective context engineering requires orchestrating multiple search tools across local files, databases, web, and memory rather than relying on single-vector retrieval.
Shell tool versatility
CLI-based tools (bash/exec) serve as universal adapters, enabling agents to navigate filesystems, execute curl commands, or generate custom scripts for any data source.
⚠️ Critical Failure Points 3 insights
The three breakdowns
Agentic search fails when agents skip tools entirely, select the wrong tool type (e.g., web vs. database), or generate invalid parameters for complex queries.
Acronym ambiguity in semantic search
Basic semantic search struggles with specific keywords and acronyms, as demonstrated when searching 'GPA' returned results about Google's Gemma models instead of GDPR talks.
Parameter complexity gradient
Simple ID lookups work reliably with small models, but free-form query languages like SQL or ESQL require significantly more capable models and careful prompting.
🛠️ Implementation Best Practices 3 insights
Invest in tool descriptions
Comprehensive descriptions must include trigger conditions, explicit 'when not to use' guidance, and tool relationships—not just single-sentence summaries.
Match model capability to tool complexity
General-purpose search tools that write entire queries from scratch demand stronger models (GPT-4o Mini) compared to simple semantic search (GPT-4o Nano).
System prompt reinforcement
When tool descriptions prove insufficient, explicitly codifying tool selection logic in system prompts resolves routing confusion between similar tools.
Bottom Line
Effective agentic search requires curating a diversified toolkit of specialized search methods with exhaustive tool descriptions, rather than expecting a single retrieval method to handle all context engineering challenges.
More from AI Engineer
View all
The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks
Sandipan Bhaumik from Databricks presents a battle-tested five-pillar framework for deploying enterprise AI agents, arguing that starting with model selection leads to inevitable production failures while proper evaluation, observability, and data governance determine success at scale.
Sovereign Escape Velocity: Ownership w Open Models — Gus Martins, & Ian Ballantyne, Google DeepMind
Google DeepMind's Gus Martins and Ian Ballantyne introduce Gemma 4, a family of open models (2B to 31B parameters) that deliver frontier-level intelligence with disproportionate efficiency, enabling sovereign AI ownership through local deployment, Apache 2.0 licensing, and on-device capabilities.
LLM Observability, Evaluation, Experimentation Platform — Dat Ngo, Arize
Dat Ngo from Arize AI explains how modern AI systems require reimagined observability and evaluation patterns built on OpenTelemetry to manage non-deterministic agents, emphasizing that the future of AI engineering lies in automated experimentation flywheels that eliminate manual dashboard work.
Text Diffusion — Brendon Dillon, Google DeepMind
Google DeepMind researcher Brendon Dillon explains text diffusion as a parallel alternative to autoregressive language models that iteratively denoises random tokens rather than generating sequentially, offering significantly lower latency and unique capabilities like self-correction and adaptive computation, though currently limited by high serving costs for large batches.