On SWE-QA, semantic search beat deep agentic search in accuracy (65.2% vs 46.2% Pass) and in cost per correct answer, with 41.8% of deep agent failures attributed to planner-to-subagent coordination breakdowns.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Deep Agentic Search for Repository-Level Code Question Answering: An Empirical Study
On SWE-QA, semantic search beat deep agentic search in accuracy (65.2% vs 46.2% Pass) and in cost per correct answer, with 41.8% of deep agent failures attributed to planner-to-subagent coordination breakdowns.