The Impact of Generative AI on Collaborative Open-Source Software Development: Evidence from GitHub Copilot
Open accessThe DOI matches its Crossref record; 147 of the 147 references it cites were found in a registry or an index; 9 of the 65 sources its claims cite were checked for support.
Abstract
Generative artificial intelligence (AI) facilitates content production and enhances ideation, with potentially important implications for developer productivity and participation in software development. To explore its impact on collaborative open-source software (OSS) development, we investigate the role of GitHub Copilot, a generative AI pair programmer, in OSS development where multiple distributed developers voluntarily collaborate. Using GitHub's proprietary Copilot usage data, combined with public OSS project data obtained from GitHub, we find that Copilot use increases project-level code contributions by 5.9%. This gain is accompanied by a 3.4% increase in developer coding participation and a 2.1% increase in individual code contributions. However, Copilot use is also associated with an 8% increase in coordination time and more code discussions. This reveals an important tradeoff: While AI expands who can contribute and how much they contribute, it slows coordination in collective development efforts. Despite this tension, the overall effect remains positive, resulting in a net increase in the timely merge of code contributions at the project level. Interestingly, we also find heterogeneous effects across developer roles. Peripheral developers exhibit relatively smaller increases in project-level code contributions and larger increases in coordination time than core developers. Together, our findings highlight the dual effects of AI pair programmers on code contributio…
Overview
CoChat Summary
TL;DR
Using proprietary GitHub Copilot usage data and public GitHub project data, this paper finds that Copilot use increases project-level code contributions but also lengthens coordination time in open-source software development. The net effect on timely merged contributions remains positive, with heterogeneous effects across core and peripheral developers.
Key findings
- Copilot use is associated with a 5.9% increase in project-level code contributions (merged PRs) and an 8% increase in coordination time (average time to merge a PR).
- Mechanism analyses show a 3.4% increase in developer coding participation, a 2.1% increase in individual code contributions, a 6.5% increase in code discussion volume per merged PR, a 1.7% increase in discussion participation, and a 5.9% increase in discussion intensity per developer.
- Despite longer coordination time, Copilot use is associated with a 3.5% increase in PRs merged within one day, 4.1% within three days, and 5.1% within ten days.
- Core developers experience a relatively larger increase in project-level code contributions (core share rises by 0.054, or 6.5%), while peripheral developers experience a larger increase in coordination time (peripheral merge-time ratio rises by 0.085, or 5.4%). Absolute changes: core contributions +6.1%, core coordination time +6.7%; peripheral contributions +5.4%, peripheral coordination time +5.3%.
Methods
- Project-month panel of GitHub OSS projects from January 2021 to December 2022; final main sample includes 7,637 projects (4,491 treatment, 3,146 control) and 139,329 observations. Treatment is defined as projects where Copilot was supported by local IDEs and used by developers, identified using proprietary GitHub Copilot usage data plus public IDE information.
- Main estimation uses the Generalized Synthetic Control Method (GSCM) with project and month fixed effects, log-transformed outcomes, and robust standard errors clustered at the project level.
- Robustness and supplementary analyses include within-IDE analysis, PSM combined with difference-in-differences, instrumental variable estimation, alternative measures, outlier removal, non-AI topic exclusion, workload controls, two-way fixed effects, expanded samples, and LDA text analysis of 5,404,735 merged PR comments.
Limitations
- The paper states that privacy constraints prevent observing individual-level Copilot usage, limiting direct measurement of heterogeneous AI adoption, role transitions, and spillovers; the observational data also cannot fully capture unobserved developer characteristics such as motivation, ability, or work preferences.
- The paper states that the dataset does not identify intermediate mechanisms through which AI tools alleviate or exacerbate coordination bottlenecks, and the panel size does not allow assessing long-term effects as developers gain experience.
- The paper states it does not examine broader social and governance implications, such as authorship transparency, credit assignment, skill erosion, bias, or security vulnerabilities; evident from the design, treatment is measured at the project level and the sample is restricted to active projects disclosing IDE information, so findings may not generalize to all OSS projects or individual Copilot use patterns.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
8 Valid · 2 Not valid · 30 Not checked · 223 Own results and interpretations
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Increased participation by developers in code discussion can introduce more perspectives, thereby extending the time required to reach consensus (Weber 2006, Feri et al. 2010, Yu et al. 2015, Roels and Corbett 2024).”
What the sources say
- Not checked — no full text of the sourcecites Weber 2006 source
- Could not verifycites Feri et al. 2010 sourceExperimental — accuracy not yet measured
The experimental instructions in the supplementary material (A1. WL-BASE—Teams-treatment; no page given) say that three-person teams could exchange messages for two minutes to agree on a decision. A further instruction allows another attempt if their entries differ. Neither passage addresses whether greater participation introduces more perspectives or lengthens the time to consensus; searches for those relationships found no passage that does. (LLM, conf=0.72)
- Not matchedcites Yu et al. 2015
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“while reducing expected costs by lowering the time and effort required to complete development tasks.”
- Also stated in: § 3.1 AI Pair Programmers and Project-level Code Contributions · p. 10 · plain
“by reducing the time and effort required to make contributions (Peng et al. 2023), AI pair programmers lower the expected costs of contributions”
What the sources say
- Partly supportedcites Peng et al. 2023 sourceExperimental — accuracy not yet measured
The Introduction reports that developers using GitHub Copilot completed a JavaScript programming task 55.8% faster than controls. The Discussion says extrapolating that productivity gain could imply cost savings, but the study did not measure effort or the costs of code contributions, and cautions that benefits may vary across tasks. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 3.1 AI Pair Programmers and Project-level Code Contributions · p. 10 · plain
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“We believe the first component satisfies the relevance condition because prior research shows that individuals learn from peers through social interactions and shared experience (Foster and Rosenzweig 1995, Conley and Udry 2010, Bollinger and Gillingham 2012, Bailey et al. 2022).”
What the sources say
- Not checked — no full text of the sourcecites Foster and Rosenzweig 1995 source
- Supportedcites Conley and Udry 2010 source
The citation supports the sentence’s stated premise: the Introduction (p. 5) reports that farmers adjust fertilizer use in response to their information neighbors’ outcomes. The paper provides evidence of learning from peers; the citing authors’ assessment of their own instrumental variable’s relevance is not a finding attributed to this source. (LLM, conf=0.92)
- Not checked — no full text of the sourcecites Bollinger and Gillingham 2012 source
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As shown in Table I.1, Copilot increased the diversity of discussion topics per merged PR”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and a negative correlation between coordination time and project-level timely merge of code contributions.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
163 more claims
Create a free account to see all 263 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 28
- It may state two separate claims as one. 46
- The quote only partly states this claim. 28
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“GitHub Copilot, for instance, has shown tangible productivity benefits in open-source environments, with studies reporting up to a 6.5% increase in project output [7].”
From Redefining the Programmer: Human-AI Collaboration, LLMs, and Security in Modern Software Engineering (2025) · cites this paper as [7]
Experimental — accuracy not yet measured
Citations
Cites
Mostly Harmless Econometrics: An Empiricist's Companion
Angrist, J. D., & Pischke, J.-S. (2009) · doi:10.1515/9781400829828 · Cited by 9,467
Griffiths, T. L., & Steyvers, M. (2004) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0307752101 · Cited by 6,049
Abadie, A., Diamond, A., & Hainmueller, J. (2010) · Journal of the American Statistical Association · doi:10.1198/jasa.2009.ap08746 · Cited by 5,513 · cited 1×
“This approach combines the idea of a synthetic control (Abadie et al. 2010) with interactive fixed effects (Bai 2009).”
MethodMotivational determinants of risk-taking behavior.
Atkinson, J. W. (1957) · Psychological Review · doi:10.1037/h0043445 · Cited by 3,706
The interdisciplinary study of coordination
Malone, T. W., & Crowston, K. (1994) · ACM Computing Surveys · doi:10.1145/174666.174668 · Cited by 3,504
Technology acceptance model: a literature review from 1986 to 2013
Marangunić, N., & Granić, A. (2015) · Universal Access in the Information Society · doi:10.1007/s10209-014-0348-1 · Cited by 2,458
Comparative Politics and the Synthetic Control Method
Abadie, A., Diamond, A., & Hainmueller, J. (2015) · American Journal of Political Science · doi:10.1111/ajps.12116 · Cited by 2,406
Some Simple Economics of Open Source
Lerner, J., & Tirole, J. (2002) · The Journal of Industrial Economics · doi:10.1111/1467-6451.00174 · Cited by 2,355
Comparative Politics and the Synthetic Control MethodPreprint
Abadie, A., Diamond, A., & Hainmueller, J. (2014) · SSRN Electronic Journal · doi:10.2139/ssrn.1950298 · Cited by 2,217
Learning about a New Technology: Pineapple in Ghana
Conley, T. G., & Udry, C. R. (2010) · American Economic Review · doi:10.1257/aer.100.1.35 · Cited by 2,180
Research works printed but not linked: 8
- J Bai, 2009, Panel data models with interactive fixed effects, Econometrica, 77, 1229, 10.3982/ECTA6135 · DOI · Paper page
- I-H Hann, 2013, All are not equal: An examination of the economic returns to different forms of participation in open source software communities, Information Systems Research, 24, 520, 10.1287/isre.2013.0474 · DOI · Paper page
- J Howison, 2014, Collaboration through open superposition: A theory of the open source way, Mis Quarterly, 38, 29, 10.25300/MISQ/2014/38.1.02 · DOI · Paper page
- R Krishnamurthy, 2016, Peripheral developer participation in open source projects: an empirical analysis, ACM Transactions on Management Information Systems (TMIS), 6, 1, 10.1145/2820618 · DOI · Paper page
- P K Medappa, 2019, Does superposition influence the success of FLOSS projects? An examination of open-source software development by organizations and individuals, Information Systems Research, 30, 764, 10.1287/isre.2018.0829 · DOI
- P Setia, 2012, How peripheral developers contribute to opensource software development, Information Systems Research, 23, 144, 10.1287/isre.1100.0311 · DOI · Paper page
- J Tsay, 2014, Let's talk about it: evaluating contributions through discussion in GitHub, Proceedings of the 22nd ACM SIGSOFT international symposium on foundations of software engineering, 144, 10.1145/2635868.2635882 · DOI · Paper page
- Y Yu, 2015, Wait for it: Determinants of pull request evaluation latency on github, IEEE/ACM 12th working conference on mining software repositories, 367, 10.1109/MSR.2015.42 · DOI · Paper page
Cited by
Who is using AI to code? Global diffusion and impact of generative AI
Daniotti, S., Wachs, J., Feng, X., & Neffke, F. M. H. (2026) · Science · doi:10.1126/science.adz9311 · Cited by 95
Redefining the Programmer: Human-AI Collaboration, LLMs, and Security in Modern Software Engineering
Cruz, E. D. L., Le, H., Meduri, K., Nadella, G. S., & Gonaygunta, H. (2025) · Computers, Materials & Continua · doi:10.32604/cmc.2025.068137 · Cited by 9
Wiesenfeld, B. M., & Kellogg, K. C. (2025) · Strategic Organization · doi:10.1177/14761270251408580 · Cited by 15
“…decision support, content generation, and prototyping and design, which allow organizations to use GenAI to increase domain knowledge around new products and services (e.g. Brynjolffson et al., 2025; Csaszar et al., 2024 Cui et al., 2024; Dell’Acqua et al., 2023; Song et al., 2024).”
Huang, A., Huang, N., & Hong, Y. (2026) · Information Systems Research · doi:10.1287/isre.2024.1551 · Cited by 2
Xu, F., Medappa, P. K., Tunc, M. M., Vroegindeweij, M., & Fransoo, J. C. (2025) · arXiv · doi:10.48550/arxiv.2510.10165 · Cited by 27
Feng, Z., Milewicz, R. M., Murphy-Hill, E. R., Menezes, T., Serebrenik, A., Steinmacher, I., & Sarma, A. (2026) · ACM Transactions on Software Engineering and Methodology · doi:10.1145/3789210 · Cited by 7
“[53].”
Impact of Generative AI Assistants on Software Engineers' Skills, Workflow Efficiency, and Decisions
Le, H. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-8450-4.ch001 · Cited by 1
Beyond the Commit: Developer Perspectives on Productivity with AI Coding Assistants
Chen, V., He, J., Williams, B., Valentino, J., & Talwalkar, A. (2026) · IEEE/ACM International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP) · doi:10.1145/3786583.3786848 · Cited by 13
"Word-of-AI" and Matching Quality: Evidence from a Natural Experiment on Online Review PlatformsPreprint
Cheng, Y., Gao, B., Zhang, R. (., & Li, X. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5052570 · Cited by 2
Assessing Open Source Software Developer's Opportunity Costs in a Time of Generative AIPreprint
Li, X., Costello, F. J., Kim, K., & Zhang, X. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5100609 · Cited by 4
The Shielding Effect of Personalization: Algorithmic Fairness in Biased Online PlatformsPreprint
Chen, Y., & Jiang, W. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5379350
Pinheiro, F. V. D. S., Coutinho, E. F., da Silva, M. M., de Carvalho, S. A. L., & Andrade, R. M. C. (2025) · Simpósio Brasileiro de Engenharia de Software (SBES) · doi:10.5753/sbes.2025.11601
RescueLens: LLM‐powered triage and action on volunteer feedback for food rescue
Raman, N. J., Tang, J., Chen, Z., Shi, Z. R., Hudson, S., Kapoor, A., & Fang, F. (2026) · AI Magazine · doi:10.1002/aaai.70080
African Migration in the Age of Surveillance and Digital Belonging
Adeniyi, A. (2026) · Springer eBooks · doi:10.1007/978-3-032-17227-3_10
Engineering Students' Usage and Perceptions of GitHub Copilot in Open-Source Projects
Rani, N., Munnangi, J. R., Spangler, A. M., & Honeycutt, D. R. (2026) · ASEE Annual Conference & Exposition · doi:10.18260/1-2--59352 · Cited by 2
de Oliveira, L. A. F., & de Oliveira, C. C. (2026) · Research Society and Development · doi:10.33448/rsd-v15i4.50966 · Cited by 1
Adoption of AI Assisted Coding – Cognitive Barriers, Motivational Factors and Productivity Gains
Varanasi, D. S., Reddy, Y. D., Potdar, S., & Warudkar, H. (2026) · Innovations in Software Engineering Conference (ISEC) · doi:10.1145/3796563.3796601 · Cited by 2
“However, these productivity gains come at a cost; for example, an observed 41.6% increase in integration times, suggesting the tool introduces additional coordination times, influencing collaborative workflows [21].”
InfluentialStrategic Release and Co-creation: Empirical Insights for Managing Open-Source Software
Chen, W., Liu, X., Krishnan, V., & Zhu, K. (2026) · Information Systems Research · doi:10.1287/isre.2021.0601
Horvat, M., Ursić, I., & Krmpotić, K. (2026) · Electronics · doi:10.3390/electronics15132805
Human-Centered AI for e-Collaboration: Software Development, Design, and Implementation
Le, H. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-8217-3.ch003
The Dose Makes the Agent: Therapeutic Index Analysis of AI Coding Contributions
Destefanis, G., Santos, R. D. S., Ortu, M., & Wessel, M. S. (2026) · International Conference on Mining Software Repositories (MSR) · doi:10.1145/3793302.3793600 · Cited by 2
Pădurean, V.-A., Riegel, K., Barbre, G., Blake, M., Denny, P. C., Gotovos, A., Leinonen, J., MacNeil, S., Prather, J. E., & Singla, A. K. (2026) · ACM Conference on International Computing Education Research (ICER) · doi:10.1145/3765964.3811639 · Cited by 1
“The skills required by computing professionals are evolving as AI becomes increasingly integrated into the software development process [11, 19, 85].”
Fardani, P. R., & Mauritsius, T. (2026) · International Conference on Applying New Technology in Green Buildings (ATiGB) · doi:10.1109/atigb70203.2026.11628312
Shan, G., Rivera, M., Kumar, S., & Anand, P. (2026) · Journal of Management Information Systems · doi:10.1080/07421222.2026.2692282 · Cited by 4
“Song et al. show that GitHub Copilot alters open-source contribution volume and review effort [79], while Pandey et al. report improved task speed and fewer bugs, though cognitive mechanisms remain unexamined [64].”
How Does Core Contributor Disengagement Impact Open Source Project Activity? A Quasi-Experiment
Chen, Y., Stol, K.-J., Santos, F., Germán, D. M., & Trinkenreich, B. (2026) · ACM/IEEE International Conference on Software Engineering: Companion Proceedings (ICSE-Companion) · doi:10.1145/3744916.3773135 · Cited by 3
Metadata
- Authors
- Song, F., Agarwal, A., & Wen, W.
- Affiliations
- The University of Texas at Austin (US)
- Corresponding author
- Wen, Wen
- Venue
- SSRN Electronic Journal (preprint)
- Publisher
- Elsevier BV
- Published
- 2024
- ISSN
- 1047-7047 (ISSN-L)
- OpenAlex
- W4400488941
- Semantic Scholar
- 712171098cc0bf2280fdf0cec1d803d6db05e18f
- Cited by
- 30
- References
- 59
- Citations by year
- 2026: 24 · 2025: 6
- FWCI
- 213.93
- Type
- journal article
- Language
- en
- Open access
- green
- doi.org
- Locations
- Information Systems Research (publishedVersion)
- SSRN Electronic Journal (acceptedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Computer Science Applications › Open Source Software Innovations
- Physical Sciences › Computer Science › Information Systems › Software Engineering Research
- Physical Sciences › Computer Science › Information Systems › Software Engineering Techniques and Practices
- Keywords
- generative AI, GitHub Copilot, open source software development, collaborative software development, code contribution, developer engagement, developer roles, core developers, open source communities, developer productivity
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.