Source-linked AI summary
Open Source Stewardship Communities: "We need you, but not your pull request"
Gregorio Robles, Daniel M. German
TL;DR
AI makes implementation cheaper while reviewing external contributions remains comparatively costly, and it is unclear whether agent-supported contribution and review preserve OSS communities’ human functions. The paper examines projects that restrict external implementation while retaining broader participation, describing this as a stewardship community. It argues that human oversight of agents does not ensure community renewal, as implementation labor may be replaced without replacing succession or diversity.
Problem
AI lowers implementation costs faster than review costs, and whether agent-supported contribution and review preserve OSS community functions remains unclear.
Source: §1 · p. 1Method
The paper examines projects that restrict external implementation while maintainers use agents, and describes the resulting arrangement as a stewardship community.
Source: §2 · pp. 2–3Results
AI can substitute for implementation labor without substituting for succession or the diversity that open contribution brought to the software and its future stewards.
Source: §4 · pp. 3–4Takeaways & Limitations
Projects may meet human-oversight goals for agents while losing the community’s direct pathway into implementation and the core, making succession and diversity explicit governance concerns.
Source: §4 · p. 4Succession and the community’s long-term capacity to sustain software may take years to become visible, much slower than immediate productivity gains.
Source: §4 · p. 4
Abstract
from arXiv · showhide
Human-centric AI for software engineering means keeping humans responsible for work performed with AI. In Open Source Software (OSS), AI lowers the cost of implementing changes, but reviewing someone else's contribution remains comparatively expensive, so some projects now restrict who may contribute implementations while still welcoming other participation---not because the code is AI-generated, but because it no longer justifies the review cost. We call the resulting form a stewardship community: a small core retains implementation authority while a broader community continues to shape the software without writing code, and access to coding increasingly depends on approval rather than self-initiated contribution. This raises a broader question: what happens to the human community when coding agents let maintainers replace implementation work once supplied by external contributors? For human-centric software engineering, keeping humans in control of AI agents is not enough: AI can replace implementation labor while weakening how OSS communities renew themselves.
1 INTRODUCTION
OSS licensing grants rights to use and modify software, but not a right to participate; contribution workflows have nevertheless let participants grow into project roles. As AI lowers implementation costs faster than review costs, some projects restrict external code while keeping other participation open, raising questions about community renewal.
1 INTRODUCTION
- Repeated contribution and review let participants gain project knowledge, demonstrate commitment, and gradually assume responsibility, producing both software and people able to sustain it.OSS licensing permits use, modification, and redistribution, but does not guarantee participation or acceptance of submitted changes.Source: §1 · p. 1
- AI lowers implementation effort faster than the cost of evaluating external changes, creating a review bottleneck as pull request volume rises while review capacity stays flat.Maintainers must understand intent, assess correctness and fit, integrate changes, and accept responsibility for maintaining them.Source: §1 · p. 1
- AI may be used on both sides of the contribution pipeline, but whether agent-supported contribution and review preserve community functions remains unclear.The potential reduction in integration burden does not resolve the question of how contribution and review sustain human participation.Source: §1 · p. 1
- Some projects close external implementation while still inviting reports, requirements, documentation, and discussion, concentrating coding within the development core.These restrictions are not necessarily responses to AI-written code; maintainers may use agents themselves to implement changes based on community knowledge.Source: §1 · p. 1
- The paper examines how restricting external implementation affects community participation, maintenance, and the pathway from user to contributor to maintainer.Its central question is whether OSS communities can renew themselves when maintainers no longer depend on outsiders’ contributions.Source: §1 · pp. 1–2
2 STEWARDSHIP COMMUNITY: GATING WHO CAN CODE
Stewardship communities retain a small core’s authority over implementation while leaving broader participation in shaping the software open. The section describes different gates to coding, examples of the pattern, and the challenge of deliberately developing future stewards.
2 STEWARDSHIP COMMUNITY: GATING WHO CAN CODE
- Contribution and review can validate code while helping contributors learn project conventions and allowing maintainers to assess their judgment, responsiveness, and fit.Review effort can therefore bring useful code, technical oversight, and progression toward maintainership.Source: §2 · p. 2
- Agentic development can make implementation cheaper for maintainers than reviewing an equivalent external contribution, while outsiders can still contribute problems, context, and desired behavior.That cost difference can reduce projects’ reliance on external pull requests for implementation.Source: §2 · p. 2
- Table 1 and five project examples illustrate maintainers using agents while restricting external implementation and retaining other forms of participation.The examples include invitation-only implementation, closed public pull requests, team-only contributions, and an issue-first workflow.Source: Table 1 · p. 2, §2 · p. 2
- Ghostty requires a maintainer’s vouch before a first pull request, then lets approved contributors submit later pull requests without seeking approval each time.This creates a persistent approved-contributor group between maintainers and the wider community; Codex’s policy does not clarify whether invitations persist.Source: §2 · p. 2
- A stewardship community separates implementation authority from broader participation: a small core decides what is implemented and merged, while others shape the resource through reports, requirements, and discussion.The gate may exclude unsolicited implementations, admit selected contributors, or pre-approve contributors or contributions; in each case, authority remains with the core.Source: §2 · p. 3
- Gating moves entry into implementation behind approval or invitation, so participation alone no longer creates an open path toward implementation, maintainership, and stewardship.The paper argues that long-term sustainability depends on how projects develop future maintainers once entry is controlled by existing members.Source: §2 · p. 3
3 OPEN QUESTIONS OF A STEWARDSHIP COMMUNITY
The section considers how gating may affect responsiveness to diverse needs and the renewal of maintainers. It describes a possible bias in insider-mediated admission and warns that development activity can grow without a corresponding increase in succession capacity.
3 OPEN QUESTIONS OF A STEWARDSHIP COMMUNITY
- The long-term effects of stewardship-based OSS models remain uncertain, though questions about diversity, responsiveness, and succession are already emerging.The paper frames these as open questions rather than established long-term outcomes.Source: §3 · p. 3
3.1 Developer diversity and responsiveness to diverse community needs
- Open contribution let users implement needs outside core priorities, whereas gating leaves implementation of reported needs to the existing core’s attention and agent capacity.Users can still propose those needs but can no longer increase their likelihood of adoption by supplying the implementation themselves; gating thus changes which needs may enter the software.Source: §3.1 · p. 3
- Because people tend to form relationships with similar others, insider-mediated admission may reproduce existing social networks among contributors and in the maintainer core.The paper identifies Ghostty’s one-time vouch requirement as an example of this admission mechanism.Source: §3.1 · p. 3
3.2 Maintainership succession
- A small agent-supported core may sustain or increase development activity while project knowledge and decision authority remain concentrated among very few people.The project may appear sustainable in the short term even as development capacity grows without corresponding succession capacity.Source: §3.2 · p. 3
- Long-term viability depends on continued software evolution, renewal of those who sustain it, and the broader community’s ability to shape its future.The consequences of gating extend beyond fewer external pull requests to who enters the development group and how future maintainers emerge.Source: §3.2 · p. 3
4 CONCLUSIONS
The conclusion argues that agentic implementation can replace outsiders’ coding labor without replacing contribution’s role in succession and diversity. It treats those losses as governance concerns while rejecting unrestricted pull-request reopening as a solution.
4 CONCLUSIONS
- AI can substitute for implementation labor without substituting for succession or the diversity open contribution brought to software and its future stewards.The shift is not primarily a reaction to AI-generated code; it reflects a reorganization of who may contribute implementation work.Source: §4 · pp. 3–4
- A project may retain human oversight of agents while losing the community’s pathway into implementation and the core, making succession and diversity governance concerns.Human oversight and community renewal are different achievements.Source: §4 · p. 4
- Reopening unrestricted external pull requests would restore the review and integration costs that made gating attractive, while its long-term consequences may take years to appear.The conclusion argues that losses from closing the pathway should be managed deliberately rather than assumed to resolve themselves.Source: §4 · p. 4