rag retrieval strategies
5 articles · 15 co-occurring · 1 contradictions · 0 briefs
Retrieved documents (Layer 3) are positioned as a distinct context layer; this implies RAG as a context engineering problem, not just a search problem.
Author argues against filesystem abstractions (like AGENTS.md as 'memory') and for direct database access with SQL. This contradicts simplified RAG patterns and suggests retrieval should be handled by the system layer, not the harness.
Retrieved documents (Layer 3) are positioned as a distinct context layer; this implies RAG as a context engineering problem, not just a search problem.
Agent traces could become a new retrieval source: 'retrieve similar agent interactions' to improve decision-making. This is RAG applied to behavioral patterns.
This implies that coding AI systems need excellent RAG for architectural information—being able to retrieve relevant dependent files without hallucinating scope.
Trace datasets function as retrieval-augmented training data for agents; the sanitization step mirrors data quality concerns in RAG pipelines.
Author argues against filesystem abstractions (like AGENTS.md as 'memory') and for direct database access with SQL. This contradicts simplified RAG patterns and suggests retrieval should be handled by