2026/03/03 by Daniel S. Katz, Warrick H. Ball, Patrick Diehl +3 · 1 voice
Computer Science · Decision Sciences · #Open Source Software Innovations #Research Data Management Practices #Scientific Computing and Data Management
paper · doi:10.5281/zenodo.18849550
openalex publication_date 2026/03/03 · openalex created_date 2026/03/04 · openalex updated_date 2026/07/15
The Journal of Open Source Software (JOSS; https://joss.theoj.org) publishes short articles describing open-source research software, with over 3200 papers/software packages published since 2016. In this talk, we aim to showcase JOSS as an open source community and platform, to encourage its use by RSEs, and to discuss how JOSS considers GenAI contributions. JOSS's platform is hosted on GitHub and incorporates scholarly community elements and open-source software that we have developed and adapted. The JOSS review process is open, iterative and includes both the submitted paper and the submitted software itself, thereby encouraging good software practice. The first aim of this talk is to discuss JOSS as an open-source community and platform. JOSS has published articles authored by RSEs and is one avenue through which to recognise RSE contributions to research software in formal research literature. This talk's second aim is to improve RSEs' awareness of JOSS for their research software projects and encourage them to engage with platforms like JOSS as authors, reviewers or even editors. We will also discuss how generative AI (GenAI) is changing what JOSS publishes. From the start, JOSS's goal was to create a relatively simple way for developers of research software to be credited for the work using the scholarly publishing system. Our initial submission criterion was simply that we only reviewed and published what we defined as research software, not more general open-source software. Five years ago, we introduced a second criterion, "substantial scholarly effort", to ensure that we published only meaningful contributions to research software, since other avenues such as Zenodo allowed for publishing any contributions, and we wanted to align a JOSS publication with similar amounts of work in other journals. Our benchmark at the time – at least roughly three months of developer time – offered a practical, human-centered proxy for effort. For that era, it served us well. Today, the landscape is fundamentally different. Where generating code once took three months, in some cases this can now be accomplished in three days – or even three hours – with generative AI tools. With AI agents increasingly capable of producing entire codebases and documentation from natural language prompts, the marginal cost of research software generation is being reduced, although the quality of the generated work may vary widely. On the positive side, this may enable people from previously underrepresented backgrounds to create new software tools. But it also poses a fundamental challenge: how do we evaluate what constitutes credit-worthy, publishable research software contributions? The third aim of the talk is to share some of JOSS's thinking about how to respond to GenAI and get feedback from the deRSE attendees.