AI coding agents can maintain or improve code quality when managed through proper quality layers: spec-driven development with AI review, high-coverage unit tests via test-driven development, manual testing, and extensive end-to-end tests. With this approach, teams can achieve 2-3x productivity gains while reducing bugs.
Researchers identified defects or ambiguous requirements in 37 of DeepSWE's 113 tasks (32.7%), a benchmark used to evaluate AI models like GPT-6 Astra and Fable 5. Issues included hidden tests causing build failures, assertions rejecting valid output, and unspecified requirements. Fixing confirmed defects raised measured pass rates by 4–6 percentage points, raising questions about benchmark reliability.
This paper addresses fundamental gaps in agentic software engineering: the requirement gap (requirements approximate stakeholder intent) and model gap (deployment models approximate real environments). The authors propose an assurance-revision loop using deployment evidence to iteratively narrow these gaps, framing the problem as resource allocation between human judgment and agent capability.
OpenSpec is a lightweight, configurable framework for creating and managing software specifications that keeps teams and AI coding agents aligned during development. It helps capture requirements, validate them, and verify implementations match specifications.
reqlan is a semantic engineering toolset that lets developers write requirement graphs alongside code using a domain-specific language, enabling better system specifications, static analysis, and integration of compliance and testing into codebase documentation. Built in Rust, it aims to reduce AI context tokens, prevent documentation drift, and make system intent explicit and compilable through file references, cross-linking, and structured requirement blocks.