Agents Can Reason. They Still Can't Search

5 pointsposted 6 hours ago
by askhn1234_12

1 Comments

frank1677

5 hours ago

Good 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.