3 ms·
Agents Can Reason. They Still Can't Search
- frank1677 5mo agoGood read The interesting part is that every layer independently reinvented retrieval: RAG for documents tool routing for MCP/actions skills/workflow loading context compaction/navigation for long-context agents Different abstractions, same core problem: “How does the agent find the right thing at the right time under cost/latency constraints?” Feels like search is becoming the actual runtime for agents.