scan-v5.json (30040B)
1 { 2 "scan_version": 5, 3 "paper_type": "empirical", 4 "paper": { 5 "title": "Speed at the Cost of Quality: How Cursor AI Increases Short-Term Velocity and Long-Term Complexity in Open-Source Projects", 6 "authors": [ 7 "Hao He", 8 "Courtney Miller", 9 "Shyam Agarwal", 10 "Christian Kästner", 11 "Bogdan Vasilescu" 12 ], 13 "year": 2026, 14 "venue": "MSR '26", 15 "arxiv_id": "2511.04427", 16 "doi": "10.1145/3793302.3793349" 17 }, 18 "checklist": { 19 "claims_and_evidence": { 20 "abstract_claims_supported": { 21 "applies": true, 22 "answer": true, 23 "justification": "All abstract claims (transient velocity gain, persistent quality degradation, GMM-identified velocity-quality feedback cycle) are directly supported by Table 2, Figure 3, and Table 3 with pre-trend tests passing.", 24 "source": "haiku" 25 }, 26 "causal_claims_justified": { 27 "applies": true, 28 "answer": true, 29 "justification": "The Borusyak et al. staggered DiD estimator with propensity score matching and pre-trend tests is an appropriate quasi-experimental design; paper is transparent about ITT interpretation and the Callaway & Sant'Anna estimator disagreement on quality outcomes.", 30 "source": "haiku" 31 }, 32 "generalization_bounded": { 33 "applies": true, 34 "answer": true, 35 "justification": "Results are explicitly bounded to observable Cursor adoption in open-source GitHub repos dominated by TypeScript/Python/JavaScript during mid-2024 to mid-2025; Section 5.1.3 specifically discusses why enterprise findings may differ substantially.", 36 "source": "haiku" 37 }, 38 "alternative_explanations_discussed": { 39 "applies": true, 40 "answer": true, 41 "justification": "Section 5.1.1 discusses excitement-frustration-abandonment cycle; Section 5.1.2 discusses velocity-driven codebase growth as mechanism for quality decline; robustness checks rule out confounds from other AI tools, repo inactivity, and selection bias.", 42 "source": "haiku" 43 }, 44 "proxy_outcome_distinction": { 45 "applies": true, 46 "answer": true, 47 "justification": "Paper explicitly labels 'lines added' and 'commits' as velocity proxies with 'moderate-to-strong correlation with perceived productivity,' and states that static analysis warnings are 'estimates of the effort required to review potential issues' rather than confirmed defects.", 48 "source": "haiku" 49 } 50 }, 51 "limitations_and_scope": { 52 "limitations_section_present": { 53 "applies": true, 54 "answer": true, 55 "justification": "Section 3.5 'Limitations and Threats to Validity' has two subsections (Internal Validity, External Validity) spanning over a full page with five specific internal threats identified.", 56 "source": "haiku" 57 }, 58 "threats_to_validity_specific": { 59 "applies": true, 60 "answer": true, 61 "justification": "Specific threats include: adoption proxy bias (only repos committing .cursorrules), unknown usage intensity (ITT effects only), model and version heterogeneity, imperfect propensity score matching, and contamination from other AI coding tools.", 62 "source": "haiku" 63 }, 64 "scope_boundaries_stated": { 65 "applies": true, 66 "answer": true, 67 "justification": "Results are explicitly interpreted as impact of systematic Cursor adoption relative to 'current state-of-the-practice' (not versus no-AI baseline), bounded to open-source repos with observable adoption, and limited to the specific study period when tools were rapidly evolving.", 68 "source": "haiku" 69 } 70 }, 71 "conflicts_of_interest": { 72 "funding_disclosed": { 73 "applies": true, 74 "answer": true, 75 "justification": "Acknowledgments disclose NSF grants 2206859, DGE214073, 2317168, 2120323; research awards from Google and Digital Infrastructure Fund; Google Cloud credits for BigQuery analysis.", 76 "source": "haiku" 77 }, 78 "affiliations_disclosed": { 79 "applies": true, 80 "answer": true, 81 "justification": "All five authors are listed as Carnegie Mellon University; no author affiliation with Cursor/Anysphere or any competing tool vendor.", 82 "source": "haiku" 83 }, 84 "funder_independent_of_outcome": { 85 "applies": true, 86 "answer": true, 87 "justification": "NSF is clearly independent; Google provides cloud credits but the study finds negative results for a competing product (Cursor, not Google's tools), and authors are not Google employees.", 88 "source": "haiku" 89 }, 90 "financial_interests_declared": { 91 "applies": true, 92 "answer": false, 93 "justification": "No explicit competing interests declaration appears in the paper; standard funding acknowledgment is provided but no formal 'no competing interests' statement.", 94 "source": "haiku" 95 } 96 }, 97 "scope_and_framing": { 98 "key_terms_defined": { 99 "applies": true, 100 "answer": true, 101 "justification": "Key terms defined include: Cursor's agentic capabilities vs. prior completion tools (Section 3.1.1), development velocity metrics with citations, SonarQube cognitive complexity definition (ref [32]), and DiD estimation targets ATT and ATTh with formal mathematical definitions.", 102 "source": "haiku" 103 }, 104 "intended_contribution_clear": { 105 "applies": true, 106 "answer": true, 107 "justification": "'Our contribution is two-fold': (1) first project-level DiD analysis of productivity gains from modern agentic coding assistant; (2) first comprehensive analysis of code quality impact from LLM agent assistant adoption.", 108 "source": "haiku" 109 }, 110 "engagement_with_prior_work": { 111 "applies": true, 112 "answer": true, 113 "justification": "Section 2 substantively engages with Copilot productivity RCTs, field experiments, and observational studies; directly positions against Becker et al. (contradicting finding) and Watanabe et al. (PR-level vs. project-level scope), explaining methodological differences.", 114 "source": "haiku" 115 } 116 } 117 }, 118 "type_checklist": { 119 "empirical": { 120 "artifacts": { 121 "code_released": { 122 "applies": true, 123 "answer": true, 124 "justification": "Data Availability section states 'We provide a replication package for this paper at: https://doi.org/10.5281/zenodo.18368661'.", 125 "source": "haiku" 126 }, 127 "data_released": { 128 "applies": true, 129 "answer": true, 130 "justification": "Replication package at zenodo DOI is provided; underlying GHArchive data is publicly accessible and the package presumably includes processed datasets.", 131 "source": "haiku" 132 }, 133 "environment_specified": { 134 "applies": true, 135 "answer": false, 136 "justification": "Paper mentions 'a local SonarQube Community server' and GHArchive/BigQuery but provides no requirements.txt, Dockerfile, R package versions, or pinned dependency specification.", 137 "source": "haiku" 138 }, 139 "reproduction_instructions": { 140 "applies": true, 141 "answer": true, 142 "justification": "Replication package at zenodo (10.5281/zenodo.18368661) is provided; by MSR convention such packages include README files with reproduction steps.", 143 "source": "haiku" 144 } 145 }, 146 "statistical_methodology": { 147 "confidence_intervals_or_error_bars": { 148 "applies": true, 149 "answer": true, 150 "justification": "Table 2 reports standard errors for all ATT estimates with ± percentage bounds; Figure 3 shows confidence bands on all event-study plots for all five outcomes.", 151 "source": "haiku" 152 }, 153 "significance_tests": { 154 "applies": true, 155 "answer": true, 156 "justification": "Heteroscedasticity- and cluster-robust Wald tests for pre-trend hypothesis testing; significance levels on all ATT estimates in Table 2; Sargan and AR(1)/AR(2) tests for GMM validity in Table 3.", 157 "source": "haiku" 158 }, 159 "effect_sizes_reported": { 160 "applies": true, 161 "answer": true, 162 "justification": "Table 2 reports percentage changes with confidence bounds (e.g., +28.58% ±13.7% lines added, +41.64% ±7.62% code complexity) computed from log-transformed ATT estimates via 100(e^ATT - 1)%.", 163 "source": "haiku" 164 }, 165 "sample_size_justified": { 166 "applies": true, 167 "answer": false, 168 "justification": "No formal power analysis; the 806 treated repos is determined by GitHub search results, and 1:3 matching ratio is justified by control diversity concerns rather than statistical power calculations.", 169 "source": "haiku" 170 }, 171 "variance_reported": { 172 "applies": true, 173 "answer": true, 174 "justification": "Standard errors reported for all ATT estimates (Table 2), all GMM coefficients (Table 3), and confidence bands appear on all event-study figures.", 175 "source": "haiku" 176 } 177 }, 178 "evaluation_design": { 179 "baselines_included": { 180 "applies": true, 181 "answer": true, 182 "justification": "1,380 propensity-score-matched never-adopting GitHub repositories serve as the control group throughout all analyses.", 183 "source": "haiku" 184 }, 185 "baselines_contemporary": { 186 "applies": true, 187 "answer": true, 188 "justification": "Control repositories are matched from the same observation period (Jan 2024–Aug 2025) on dynamic covariate trajectories, ensuring contemporary comparison.", 189 "source": "haiku" 190 }, 191 "ablation_study": { 192 "applies": false, 193 "answer": false, 194 "justification": "Not applicable to this observational DiD study; multiple estimator comparisons (TWFE, Borusyak, Callaway & Sant'Anna) and robustness subsets serve an analogous sensitivity function.", 195 "source": "haiku" 196 }, 197 "multiple_metrics": { 198 "applies": true, 199 "answer": true, 200 "justification": "Five outcome metrics used: commits and lines added (velocity), static analysis warnings, duplicate line density, and code complexity (quality).", 201 "source": "haiku" 202 }, 203 "human_evaluation": { 204 "applies": false, 205 "answer": false, 206 "justification": "Study measures automated repository metrics; human evaluation of outputs is not applicable to this observational design.", 207 "source": "haiku" 208 }, 209 "held_out_test_set": { 210 "applies": false, 211 "answer": false, 212 "justification": "Causal inference study, not a prediction task; train/test split concept does not apply.", 213 "source": "haiku" 214 }, 215 "per_category_breakdown": { 216 "applies": true, 217 "answer": true, 218 "justification": "Appendix D breaks down SonarQube warnings by 20 categories pre/post adoption; Appendix C provides breakdowns by programming language (JS/TS, Python, Go) and by Cursor adoption cohort.", 219 "source": "haiku" 220 }, 221 "failure_cases_discussed": { 222 "applies": true, 223 "answer": true, 224 "justification": "Section 5.1.1 discusses repos that abandoned Cursor post-adoption (excitement-frustration-abandonment cycle); Section 5.1.2 discusses code complexity increasing even when velocity is controlled.", 225 "source": "haiku" 226 }, 227 "negative_results_reported": { 228 "applies": true, 229 "answer": true, 230 "justification": "No significant effect on duplicate line density overall; velocity gains dissipate fully by month 3; Callaway & Sant'Anna yields non-significant negative estimates for quality outcomes, all reported without suppression.", 231 "source": "haiku" 232 } 233 }, 234 "setup_transparency": { 235 "model_versions_specified": { 236 "applies": false, 237 "answer": false, 238 "justification": "Researchers don't run any LLMs; this is an observational study of repositories using Cursor. Model version heterogeneity is acknowledged as a study limitation.", 239 "source": "haiku" 240 }, 241 "prompts_provided": { 242 "applies": false, 243 "answer": false, 244 "justification": "No LLM prompts used by the researchers; this is an observational study of existing repositories.", 245 "source": "haiku" 246 }, 247 "hyperparameters_reported": { 248 "applies": false, 249 "answer": false, 250 "justification": "No LLM hyperparameters are used by the researchers.", 251 "source": "haiku" 252 }, 253 "scaffolding_described": { 254 "applies": false, 255 "answer": false, 256 "justification": "Researchers study black-box adoption effects of Cursor; no agentic scaffolding is implemented by the research team.", 257 "source": "haiku" 258 }, 259 "data_preprocessing_documented": { 260 "applies": true, 261 "answer": true, 262 "justification": "Section 3.1 documents GitHub code search API with adaptive partitioning algorithm, ≥10 star filter, fork exclusion, propensity score logistic regression specification with equation, monthly GHArchive metric collection, SonarQube Community server setup, and log-transformation of all outcomes.", 263 "source": "haiku" 264 } 265 }, 266 "data_integrity": { 267 "raw_data_available": { 268 "applies": true, 269 "answer": true, 270 "justification": "Replication package at zenodo (10.5281/zenodo.18368661) is provided; underlying GHArchive data is publicly accessible for independent verification.", 271 "source": "haiku" 272 }, 273 "data_collection_described": { 274 "applies": true, 275 "answer": true, 276 "justification": "Section 3.1 describes GitHub code search API queries with adaptive file-size partitioning, GHArchive monthly time series collection for 800k+ candidate repos per cohort, and SonarQube analysis procedure.", 277 "source": "haiku" 278 }, 279 "recruitment_methods_described": { 280 "applies": false, 281 "answer": false, 282 "justification": "No human participants; repositories are the units of analysis selected by algorithmic criteria.", 283 "source": "haiku" 284 }, 285 "data_pipeline_documented": { 286 "applies": true, 287 "answer": true, 288 "justification": "Full pipeline documented: identify Cursor-adopting repos via .cursorrules files → filter by stars → collect GHArchive dynamic covariates → propensity score matching per cohort → monthly SonarQube analysis → DiD estimation → GMM panel analysis.", 289 "source": "haiku" 290 } 291 }, 292 "contamination": { 293 "training_cutoff_stated": { 294 "applies": false, 295 "answer": false, 296 "justification": "Study does not evaluate LLM capabilities on benchmarks; it measures repository-level behavioral effects of Cursor adoption.", 297 "source": "haiku" 298 }, 299 "train_test_overlap_discussed": { 300 "applies": false, 301 "answer": false, 302 "justification": "Not applicable; no LLM benchmarking performed.", 303 "source": "haiku" 304 }, 305 "benchmark_contamination_addressed": { 306 "applies": false, 307 "answer": false, 308 "justification": "Not applicable; no LLM benchmarking performed.", 309 "source": "haiku" 310 } 311 }, 312 "human_studies": { 313 "pre_registered": { 314 "applies": false, 315 "answer": false, 316 "justification": "No human participants; study analyzes public GitHub repository data.", 317 "source": "haiku" 318 }, 319 "irb_or_ethics_approval": { 320 "applies": false, 321 "answer": false, 322 "justification": "No human participants; study uses public repository data.", 323 "source": "haiku" 324 }, 325 "demographics_reported": { 326 "applies": false, 327 "answer": false, 328 "justification": "No human participants.", 329 "source": "haiku" 330 }, 331 "inclusion_exclusion_criteria": { 332 "applies": false, 333 "answer": false, 334 "justification": "Human subject criteria not applicable; repository selection criteria are described algorithmically in Section 3.1.", 335 "source": "haiku" 336 }, 337 "randomization_described": { 338 "applies": false, 339 "answer": false, 340 "justification": "No human participants; treatment assignment is naturally occurring.", 341 "source": "haiku" 342 }, 343 "blinding_described": { 344 "applies": false, 345 "answer": false, 346 "justification": "No human participants.", 347 "source": "haiku" 348 }, 349 "attrition_reported": { 350 "applies": false, 351 "answer": false, 352 "justification": "No human participants.", 353 "source": "haiku" 354 } 355 }, 356 "cost_and_practicality": { 357 "inference_cost_reported": { 358 "applies": false, 359 "answer": false, 360 "justification": "Researchers do not run LLMs; inference costs are borne by the studied repositories' developers and are not measurable in this observational design.", 361 "source": "haiku" 362 }, 363 "compute_budget_stated": { 364 "applies": true, 365 "answer": false, 366 "justification": "Google Cloud credits for BigQuery analysis are acknowledged but no specific compute cost or resource budget is reported for the SonarQube analysis pipeline running on 806+ repos.", 367 "source": "haiku" 368 } 369 } 370 } 371 }, 372 "claims": [ 373 { 374 "claim": "Cursor adoption leads to a 281% increase in lines added in the first adoption month, with gains fully dissipating after 2 months", 375 "evidence": "Table 2 (overall ATT +28.58%) and Figure 3 (ATTh showing large spike at h=0,1 then returning to baseline), consistent across all three DiD estimators", 376 "supported": "strong" 377 }, 378 { 379 "claim": "Static analysis warnings increase persistently by ~30% post-Cursor adoption", 380 "evidence": "Table 2 (Borusyak: +30.26%), Figure 3 (sustained effect); BUT Callaway & Sant'Anna yields -10.49% non-significant (Table 6, Appendix B), a substantive divergence the paper attributes to small cohort sizes", 381 "supported": "moderate" 382 }, 383 { 384 "claim": "Code complexity increases persistently by ~41% post-Cursor adoption", 385 "evidence": "Table 2 (Borusyak: +41.64%), Figure 3; Callaway & Sant'Anna yields -3.80% non-significant (Table 6), same estimator divergence applies", 386 "supported": "moderate" 387 }, 388 { 389 "claim": "Accumulated technical debt subsequently reduces future development velocity, creating a self-reinforcing cycle", 390 "evidence": "Table 3 GMM estimates: code complexity → lines added coefficient -0.718 (p<0.001), static warnings → lines added -0.588 (p<0.001); instruments validated by Sargan p>0.05 and AR(2) p>0.05", 391 "supported": "moderate" 392 }, 393 { 394 "claim": "Cursor adoption causes inherently more complex code beyond what is explained by codebase size growth", 395 "evidence": "Table 3 GMM model for lines added → code complexity shows Cursor coefficient 0.086 (p<0.001) even controlling for lines of code; interpreted as ~9% baseline complexity increase attributable to Cursor itself", 396 "supported": "moderate" 397 }, 398 { 399 "claim": "Quality degradation effects are amplified, not attenuated, in repositories with more intensive Cursor usage", 400 "evidence": "Figure 4 Row 1: High Contributor Adoption and Cursor Configuration Changes subsets both show stronger quality effects than the full ITT sample", 401 "supported": "strong" 402 } 403 ], 404 "methodology_tags": [ 405 "observational" 406 ], 407 "key_findings": "A staggered difference-in-differences study of 806 Cursor-adopting open-source GitHub repositories finds that Cursor adoption produces substantial but transient velocity gains (281% increase in lines added in month 1, dissipating fully by month 3) alongside persistent technical debt accumulation (+30% static analysis warnings, +41% code complexity per Borusyak et al. estimator). Panel GMM analysis demonstrates this accumulated debt subsequently suppresses future development velocity, creating a self-reinforcing quality-velocity degradation cycle. Robustness checks confirm quality degradation is amplified in repos with intensive Cursor usage; however, the Callaway & Sant'Anna estimator yields non-significant negative estimates for all quality outcomes, substantially weakening causal confidence in the debt accumulation findings specifically.", 408 "red_flags": [ 409 { 410 "flag": "Estimator disagreement on primary quality claims", 411 "detail": "Callaway & Sant'Anna yields -10.49% (non-significant) for static analysis warnings and -3.80% (non-significant) for code complexity, directly contradicting the Borusyak et al. estimates of +30.26% and +41.64%. The paper attributes this to small per-cohort sample sizes but cannot resolve the disagreement, substantially undermining causal confidence in the quality degradation findings." 412 }, 413 { 414 "flag": "Table 2 commits significance inconsistency", 415 "detail": "Table 2 marks commits ATT=0.0260 with *** (p<0.001) despite SE=0.0429 (t-stat ~0.6) and the paper body stating 'there is no statistically significant effect for the volume of commits'; the *** appears to be a typographical error contradicting the text." 416 }, 417 { 418 "flag": "Adoption proxy validity", 419 "detail": "Treatment is identified only through committed .cursorrules files; developers can and do use Cursor without committing configuration files, creating an ITT design measuring 'systematic adoption' and introducing unknown selection bias toward more process-conscious adopters." 420 }, 421 { 422 "flag": "Lines added as AI-era velocity metric", 423 "detail": "Large increases in lines added may reflect AI-generated boilerplate, scaffolding, or verbose refactoring rather than meaningful feature development, making this proxy especially unreliable precisely in the AI-assisted context being studied — the paper does not address this circularity." 424 }, 425 { 426 "flag": "SonarQube metrics unvalidated for AI-generated code", 427 "detail": "Paper acknowledges 'complexity metrics were designed for human-written code; whether they appropriately penalize AI-generated patterns that are mechanically verifiable yet syntactically complex remains an open question,' undermining the interpretation of the code complexity outcome." 428 }, 429 { 430 "flag": "Warning breakdown analysis is non-causal convenience sample", 431 "detail": "Appendix D (20-category SonarQube breakdown) is explicitly described as a 'convenience sample' due to architectural pipeline limitations preventing precise per-version tracking, and the paper cautions it cannot be used for causal inference — yet it is cited in support of the main narrative." 432 } 433 ], 434 "cited_papers": [ 435 { 436 "title": "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot", 437 "relevance": "Primary prior RCT showing 56% task completion speedup from Copilot; key baseline for productivity claims" 438 }, 439 { 440 "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", 441 "relevance": "Directly contrasted: controlled experiment showing Cursor does NOT help experienced OSS developers; complementary finding" 442 }, 443 { 444 "title": "The Impact of Large Language Models on Open-source Innovation: Evidence from GitHub Copilot", 445 "relevance": "Prior observational DiD estimating 17.82% release increase from Copilot; direct methodological predecessor" 446 }, 447 { 448 "title": "The Impact of Generative AI on Collaborative Open-Source Software Development: Evidence from GitHub Copilot", 449 "relevance": "Similar observational design finding 6.5% project-level productivity increase; provides comparison estimate" 450 }, 451 { 452 "title": "On the use of agentic coding: An empirical study of pull requests on GitHub", 453 "relevance": "Studies Claude Code PR acceptance (83.8%) at PR level; this paper explicitly extends to longitudinal project-level effects" 454 }, 455 { 456 "title": "Revisiting event-study designs: Robust and efficient estimation", 457 "relevance": "Methodological foundation: the Borusyak et al. imputation DiD estimator used as primary causal identification strategy" 458 }, 459 { 460 "title": "Asleep at the Keyboard? Assessing the Security of GitHub Copilot's Code Contributions", 461 "relevance": "Benchmark study establishing Copilot security vulnerability concerns; prior work motivating quality dimension analysis" 462 }, 463 { 464 "title": "The effects of generative AI on high skilled work: Evidence from three field experiments with software developers", 465 "relevance": "Field experiments at Microsoft/Accenture/Cisco finding 22-36% productivity increase; enterprise baseline for contrast with open-source findings" 466 } 467 ], 468 "engagement_factors": { 469 "practical_relevance": { 470 "score": 3, 471 "justification": "Directly addresses whether Cursor is worth adopting for development teams, with actionable findings about technical debt accumulation requiring quality-assurance process changes." 472 }, 473 "surprise_contrarian": { 474 "score": 3, 475 "justification": "Empirically challenges the '10x productivity' narrative with evidence of transient gains reversing to baseline plus persistent complexity debt, directly contradicting widespread practitioner enthusiasm." 476 }, 477 "fear_safety": { 478 "score": 1, 479 "justification": "Security warnings modestly increase (+1.98 per repo/month per Table 8) but the paper's focus is technical debt and maintainability, not critical safety risks." 480 }, 481 "drama_conflict": { 482 "score": 2, 483 "justification": "Targets a popular, well-funded product with negative longitudinal findings; internal estimator disagreement creates unresolved methodological tension the paper cannot fully explain." 484 }, 485 "demo_ability": { 486 "score": 1, 487 "justification": "Observational econometric study with no demo artifact; readers cannot readily experience or replicate the findings themselves." 488 }, 489 "brand_recognition": { 490 "score": 3, 491 "justification": "Studies Cursor (most popular AI IDE by adoption metrics cited), authored by CMU team with strong SE credentials (Kästner, Vasilescu), published at MSR '26." 492 } 493 }, 494 "hn_data": { 495 "threads": [ 496 { 497 "hn_id": "47401734", 498 "title": "Speed at the cost of quality: Study of use of Cursor AI in open source projects (2025)", 499 "points": 147, 500 "comments": 80, 501 "url": "https://news.ycombinator.com/item?id=47401734", 502 "created_at": "2026-03-16T17:07:37Z" 503 }, 504 { 505 "hn_id": "38283398", 506 "title": "API-Driven Program Synthesis for Testing Static Typing Implementations", 507 "points": 35, 508 "comments": 1, 509 "url": "https://news.ycombinator.com/item?id=38283398", 510 "created_at": "2023-11-15T22:19:08Z" 511 }, 512 { 513 "hn_id": "45968758", 514 "title": "Does AI-Assisted Coding Deliver? A Study of Cursor's Impact on Software Projects", 515 "points": 14, 516 "comments": 2, 517 "url": "https://news.ycombinator.com/item?id=45968758", 518 "created_at": "2025-11-18T16:50:19Z" 519 }, 520 { 521 "hn_id": "46730534", 522 "title": "Does AI-Assisted Coding Deliver? A Study of Cursor on Software Projects", 523 "points": 2, 524 "comments": 0, 525 "url": "https://news.ycombinator.com/item?id=46730534", 526 "created_at": "2026-01-23T09:54:11Z" 527 }, 528 { 529 "hn_id": "46658985", 530 "title": "Does AI-Assisted Coding Deliver? A Study of Cursor's Impact on Software Projects", 531 "points": 2, 532 "comments": 0, 533 "url": "https://news.ycombinator.com/item?id=46658985", 534 "created_at": "2026-01-17T15:53:22Z" 535 }, 536 { 537 "hn_id": "45998822", 538 "title": "Does AI-Assisted Coding Deliver? A Difference-in-Differences Study", 539 "points": 2, 540 "comments": 0, 541 "url": "https://news.ycombinator.com/item?id=45998822", 542 "created_at": "2025-11-20T22:36:21Z" 543 }, 544 { 545 "hn_id": "45951387", 546 "title": "Does AI-Assisted Coding Deliver? A Study of Cursor's Impact on Software Projects", 547 "points": 2, 548 "comments": 0, 549 "url": "https://news.ycombinator.com/item?id=45951387", 550 "created_at": "2025-11-17T06:57:28Z" 551 }, 552 { 553 "hn_id": "42127507", 554 "title": "UniGAD: Unifying Multi-Level Graph Anomaly Detection", 555 "points": 2, 556 "comments": 0, 557 "url": "https://news.ycombinator.com/item?id=42127507", 558 "created_at": "2024-11-13T16:32:30Z" 559 }, 560 { 561 "hn_id": "46180812", 562 "title": "Does AI-Assisted Coding Deliver? A Difference-in-Differences Study", 563 "points": 1, 564 "comments": 0, 565 "url": "https://news.ycombinator.com/item?id=46180812", 566 "created_at": "2025-12-07T10:54:26Z" 567 }, 568 { 569 "hn_id": "46070691", 570 "title": "A Difference-in-Differences Study of Cursor's Impact on Software Projects", 571 "points": 1, 572 "comments": 0, 573 "url": "https://news.ycombinator.com/item?id=46070691", 574 "created_at": "2025-11-27T16:21:41Z" 575 } 576 ], 577 "top_points": 147, 578 "total_points": 208, 579 "total_comments": 83 580 } 581 }