An experience report finds that LLM-based generation of requirements from code is not yet viable, as models fail to produce non-implemented requirements, high-quality specs, or synthetic smells reliably and require extensive human supervision for quality.
Code2Req: Using generative AI to generate requirements from source code,
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
-
On the Viability of Requirements Generation From Code: An Experience Report
An experience report finds that LLM-based generation of requirements from code is not yet viable, as models fail to produce non-implemented requirements, high-quality specs, or synthetic smells reliably and require extensive human supervision for quality.