From RAG to Agentic Search
mdancho84 · x · 2026-07-19
This post asks: Is context engineering just a new name for RAG? The author answers, "Not exactly, but they are both solving the same problem: how to build the right context for an LLM."
The accompanying image breaks down the evolution into three layers:
- Traditional RAG: A fixed retrieval pipeline that searches documents first and then stuffs the results into the context window.
- Agentic RAG: The LLM is no longer a passive receiver of retrieval results; it uses tool calls to decide when and what to retrieve.
- Agentic search in context engineering: Context sources expand further to include local files, skill files, databases, the Web, long-term memory, and Shell tools. The context window contains not just system/user prompts, but also skills YAML, tool lists, history, and tool call results.
The core takeaway: RAG is a sub-module, whereas context engineering focuses on the entire context construction and orchestration system.
Related event: Context Engineering Overtakes RAG in AI Development(2 posts)→
More from coding & agent
- He benchmarks new models by having them build their own copilot — $28 of compute — bucolucas · 2026-10-27
- Should agents be allowed to change plans mid-task? Reddit debates the drift boundary — wistfulcracker84 · 2026-10-27
- HeyGen + Canva MCP turns slide speaker notes into an avatar presentation video — HeyGen · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Cloudflare Computer adds agent interruption recovery and approval pauses in v0.7.0 — irvinebroque · 2026-10-27
- LLM Tails cuts agent input tokens 80% by making agents read before they write — savadks1818 · 2026-10-27