Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
Factors Impacting Developer Efficiency: Results from an Adaptive Longitudinal Study
Citation. Ribeiro, D., Alves, B., Souza, G., França, C., & Souza, A. (2026). Factors Impacting Developer Efficiency: Results from an Adaptive Longitudinal Study (arXiv:2608.16596).
Summary. This preprint reports a mixed-methods longitudinal case study using twelve survey waves and eighteen semi-structured interviews with 27 external software developers in a Brazilian professional-training context. It distinguishes persistent organizational barriers, adaptive technical barriers, moderately stable working-condition barriers, and emergent generative-AI barriers. The adaptive monitoring instrument captured AI-use difficulties after they emerged in interviews, but the study measures self-reported perceived efficiency rather than objective productivity or causal effects.
Key claims
- Dependencies on other teams and waiting for external validation were the most frequently reported barriers, averaging 40.1% and 45.8% respectively across the waves in which they were measured, and remained structurally persistent after response-rate sensitivity checks.
- Technical knowledge gaps, time spent searching for information, and infrastructure problems declined during early waves but rebounded later, so apparent adaptation was not durable resolution.
- Generative-AI items were added only for the final three waves after AI-related difficulties emerged in qualitative interviews; non-use when AI could have helped was reported by 22.2% in each of those waves, while time validating or adjusting AI-generated code was reported by 11.1%–18.5%.
- A fixed initial survey would have missed an AI-use theme that became the most frequently coded interview theme, illustrating the value of adaptive monitoring for changing work conditions.
Evidence & limitations
- The study combines recurring self-report surveys with thematic analysis of 18 interviews, but it has no control group and cannot causally attribute changes to experience, task changes, or the intervention context.
- The sample is 27 developers in one Brazilian education-and-consulting program; findings are not statistically generalizable to other organizations, countries, or software domains.
- AI-related results cover only the final three waves and therefore do not establish long-term AI-use trends.
- Public provenance was verified against the title-page arXiv annotation: the public arXiv PDF for 2608.16596 has the same 806,465-byte binary SHA-256 as the ingested Drive PDF. The arXiv API returned HTTP 406 and Crossref rate-limited the exact-title query (HTTP 429), so no separate publisher record was used.