The integration of agentic artificial intelligence into software development workflows has introduced a new class of challenges for open-source software communities. As autonomous AI systems become capable of independently planning, implementing, and submitting code contributions, maintainers must deal with pull requests whose origin is not always disclosed and whose quality may not reflect sufficient human oversight. Despite growing community friction around this shift, evidenced by explicit AI contribution policies, controlled empirical studies of how maintainers actually respond to agentic AI contributions remain scarce.

This thesis investigates maintainer reception of agentic AI pull requests by actively submitting 90 pull requests to 45 open-source repositories across Python, TypeScript, and Java, targeting good first issues: tasks traditionally reserved for newcomers making their first contribution to a project. Contributions are structured along two dimensions: whether the repository has explicitly configured agentic AI tooling in its development workflow, and whether the use of AI assistance is disclosed in the pull request. This yields three contribution types, covering disclosed and undisclosed submissions to repositories without explicit AI configuration and disclosed submissions to repositories that have integrated agentic AI tooling. A mixed-methods approach is applied, combining quantitative analysis of acceptance rates and review activity with a qualitative thematic analysis of maintainer feedback.

The results show that acceptance rates differed across contribution types, with repositories that had explicitly integrated agentic AI tooling showing a statistically significantly lower acceptance rate compared to standard repositories with disclosed AI assistance, though not relative to the undisclosed group. Across all groups, staleness accounted for the majority of non-merged pull requests, suggesting that non-engagement was a more common outcome than active rejection. Disclosing AI assistance made no meaningful difference to acceptance rates within the same repository context. No statistically significant differences were found in the volume of reviews or comments across groups, although automated bots contributed a notable share of interactions, particularly in repositories with agentic tooling integration. Thematic analysis of maintainer feedback showed that code quality and implementation correctness were the dominant concerns across all groups, while explicit distrust of AI-generated contributions remained low. When maintainers did reject contributions on AI-related grounds, the concern was typically the degree of human oversight behind the submission rather than AI use itself. Several repositories also introduced or revised AI policies during the contribution period, reflecting how actively norms in this space are still evolving.

Behavior-agnostic reinforcement learning is a rapidly expanding research area focusing on developing algorithms capable of learning effective policies without explicit knowledge of the environment's dynamics or specific behavior policies. It proposes robust techniques to perform off-policy evaluation, namely Distribution Correction Estimation (DICE) methods, in the context of infinite horizon Markov Decision Processes (MDPs). This research paper investigates the impact of the initial start distribution mismatch on the accuracy of DICE estimators in behavior-agnostic reinforcement learning. To achieve this, seven systematic initial start distributions were created and utilized to calculate the initial start distribution mismatch via Kullback–Leibler (KL) divergence. Furthermore, off-policy evaluation performance was assessed using DICE estimators, with Mean Squared Error (MSE) comparisons against ground truth values. The study reveals that, based on the conducted experiments, the initial start distribution mismatch does not have a clear influence on the performance of the DICE estimators. Therefore, future research is required to increase the scope of the experiments and address some of the limitations of this study to accurately assess the impact of the initial start distribution mismatch on off-policy evaluation using DICE methods. This paper underscores the complexity of the initial start distribution choice in behavior-agnostic reinforcement learning, calling for further research to effectively evaluate its impact across diverse environments and measures. Additionally, exploring the relation between the initial start distribution and policies could provide deeper insights and further refine the understanding of their influence on DICE estimators.