An observational study of 90 agent runs finds that higher reasoning effort, not testing tool access, drives first-try reliability in agentic code generation.
Realtime Retrospective Board: AI Model Benchmark Dataset and Evaluation Artifacts
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Reasoning effort, not tool access, buys first-try reliability in agentic code generation: an observational study
An observational study of 90 agent runs finds that higher reasoning effort, not testing tool access, drives first-try reliability in agentic code generation.