Specialty
A review of public discussions, interviews and repositories finds that many software specifications mix requirements with implementation detail, while practitioners use the workflow selectively.
Scientific publication4 min read
A preprint study of software repositories found that coding-agent adoption coincided with more activity, but repositories with more structured committed AI practices had smaller increases in some static-analysis quality measures. The authors stress that the findings are associations, not proof that configuration caused the difference.
Scientific publication4 min read
A software gate that compared Python and SPARQL calculations agreed on all seven tests, yet three published counts were wrong.
Scientific publication3 min read
A small benchmark found that coding agents could turn detailed specifications and empty workspaces into mostly working repositories, but the last few failures often came from narrow, easy-to-miss requirements.
Scientific publication5 min read
A small observational study found that software-development tasks involving generative AI were associated with higher self-rated cognitive load, but task type mattered more and wearable data added little.
Scientific publication4 min read
A new benchmark found that coding agents spent much of their effort locating relevant code across large systems, while the best tested setup completed 26.3% of tasks.
Scientific publication5 min read
A preprint presents a symbolic-numeric eigensolver that recovered repeated solutions in a two-equation polynomial example and reconstructed a sample symmetric tensor. The reported results illustrate the method, but do not establish broad speed, stability, or accuracy.
Scientific publication4 min read
Benchmark tests on three RVV 1.0 RISC-V CPU configurations found that juFFTe's gains varied by hardware, with a reported threefold average speedup on the SG2044 multi-core comparison, while AMD Zen 5 led across the full range.
Scientific publication4 min read
Across 120 slots per condition, first post-edit re-verification appeared in 78.3% of cadence-guided slots and 26.7% of cadence-omitted slots. The same descriptive comparison showed fewer cadence violations and more bounded final successes with the guidance.
Scientific publication4 min read
A preprint presents URIUM as an open, adaptable compiler-design course built around a small language, a staged compiler and several assembly backends.
Scientific publication3 min read
A preprint proposes ARISMA, a governance and reporting framework that keeps AI assistance bounded, validated and under human control throughout an evidence review.
Scientific publication3 min read
A qualitative preprint examines one agent-built software project and six industry accounts, arguing that durable progress may require explicit models, automated checks and clear human authority.
Scientific publication4 min read
A numerical preprint tests a site-by-site map of information loss and model-based reconstruction. Its reported losses vary between uniform synthetic noise and hardware PTMs.
Scientific publication4 min read
A methods preprint found that a deterministic auditor passed its finite template test, but positive predictive scores did not translate into positive risk-adjusted performance in the historical holdout.
Scientific publication5 min read
A preprint reports that RotDroid found more true bugs than DLD and a DOC-based method under the same app-testing budget, while RotVL-8B posted higher F1 scores than GPT-5.2 on the listed benchmark tasks.
Scientific publication3 min read
A formal software framework pairs undoable effects with reactive dependency management, but its practical evidence remains limited to one ecosystem.
Scientific publication4 min read
A preprint reports seven usability factors associated with ratings of M-commerce applications in Pakistan and strong performance from a regression model built on 168 analyzed instances. The result is exploratory: the study excluded 32 instances and reports internal validation rather than an independent external test.
Scientific publication3 min read
An arXiv preprint reports that OpsHarness, an external layer around a general AI agent, recorded higher root-cause accuracy than the tested alternatives on public benchmarks and one company dataset, with similar per-case diagnosis costs.
Scientific publication4 min read
A modeling study found that keystroke-level editing features improved early separation of Breakthrough and Fully Stuck student-exercise instances, while precision remained limited when the model was set to catch 70% of Fully Stuck cases.
Scientific publication5 min read
A methods study found that specification-valid Office files could show BENIGN in the default editing canvas while extractors and AI interfaces exposed task-relevant TRAP content.
Scientific publication5 min read
A preprint reports that an identifier-renaming attack intended to preserve functionality increased query–code similarity and reduced retrieval scores in tested systems, while the reason for cross-model transfer remains unresolved.
Scientific publication5 min read
A preprint reports stronger benchmark ranking than BM25 in one evaluation, and larger proxy component counts and high self-rated usability for RaWi in a separate study of 19 participants.
Scientific publication4 min read
A preprint evaluating two automated model-card workflows found strong results for reorganizing existing cards, while generation from repository files still produced source mismatches and incorrect subsections.
Scientific publication4 min read
A qualitative study of three open-source coding-agent harnesses finds a recurring five-part architecture, while external verifiability remains absent across the cases.
Scientific publication4 min read
A computational preprint presents Opal.jl as a shared workflow for filters, variational methods and interchangeable model backends, with benchmark results that show both accurate recoveries and clear trade-offs.
Scientific publication4 min read
A methods evaluation of Metis reports faster matched tool runs in one workload, but fault, permission and maintenance tests limit what can be claimed about its safety or reliability.
Scientific publication5 min read
An evidence-using large-language-model system outperformed tested baselines on CVSS vulnerability assessment, while a small practitioner study found its supporting material useful but exposed uncertainty around incorrect predictions and reference labels.
Scientific publication4 min read
A formal study of compiler inlining reports a worst-case cost no higher than 2.5 times the model’s global optimum, while stressing that the result is a theorem about an abstract model rather than a benchmark of real software.
Scientific publication3 min read
An arXiv preprint describes expanded support for generating proof obligations for VDM operations and reports a lower “UNCHECKED” share in one corpus comparison.
Scientific publication4 min read
A preprint evaluating Cypress end-to-end testing on the open-source Conduit RealWorld application reports a large descriptive speed difference against manual execution and lower repair times for a suite using data-cy locators. The study also found that a minority of scenarios were flaky, with instability concentrated in the Favorite Article workflow.
9th International Conference on Research in Engineering and Technology (RET 2026)4 min read
A preprint analyzing repository-local documentation activity found that coding agents most often interacted with instruction files and working notes. It found no consistent link between consultation and code editing or documentation creation, while testing and building appeared less frequent afterward.
Scientific publication4 min read
BreakGuard’s strongest reported setup detected 27 of 89 dependency-breaking updates, but generated tests often failed to compile and mainly exposed crash-type failures rather than changed values.
Scientific publication4 min read