Call for Papers

Theme for 2027: Software Architecture Meets Agentic AI

SAML, the International Workshop on Software Architecture and Machine Learning, brings together researchers and practitioners working at the intersection of software architecture and modern ML techniques, including classical ML, LLMs, GenAI, and agentic AI. The workshop covers both directions of the combination, SA4ML and ML4SA. This year, we especially welcome work on foundation-model-based, LLM-based, and agentic systems.

The theme has two complementary directions. First, how should we architect agentic AI systems whose behavior emerges from models, tools, data, memory, workflows, and runtime control mechanisms? Second, how can agentic AI techniques support software architecture activities such as design exploration, documentation, analysis, evaluation, decision-making, and evolution?

Topics of Interest

SAML 2027 welcomes work in either direction below, and especially work that connects both. While our focus in the next edition is on agentic AI, all other forms of ML and their intersection with software architecture are still very welcome as submissions.

Software Architecture for Agentic AI Systems

How should we design, evaluate, operate, and evolve systems whose behavior emerges from models, tools, memory, data, workflows, and runtime governance?

  • Architecture design, evaluation, patterns, and trade-offs for ML, LLM, and agentic systems
  • Runtime architecture for orchestration, tool integration, memory and state management, monitoring, governance, and harness engineering
  • Architecting data, ML, and LLM pipelines or workflows
  • MLOps, LLMOps, and AgentOps practices from a software architecture perspective
  • Quality attributes, technical debt, and evolution of ML, LLM, and agentic systems
  • Human, organizational, and social aspects of architecting ML-enabled systems

Agentic AI for Software Architecture

How can agentic AI, ML, and LLM techniques help architects design, analyze, document, evaluate, decide, refactor, and evolve software systems?

  • Agentic AI, ML, and LLMs for architecture design, analysis, evaluation, and decision-making
  • Agentic AI, ML, and LLMs for architecture documentation, knowledge extraction, design intelligence, and refactoring
  • Agentic workflows for architecture review, modernization, compliance, and trade-off exploration
  • Human-AI collaboration patterns for architecture work

Submission Types

We solicit two types of submissions:

Registered Reports

Detailed study designs submitted for feedback before study execution, with a clear path toward a full paper in ACM TOSEM.

Limit: 6 pages + 2 pages for references.

Short Papers

Industry experience reports, position papers, vision papers, reflection papers, and new idea papers that support workshop discussion.

Limit: 5 pages + 1 page for references.

SAML 2027 does not solicit full research papers. We aim for submissions that support workshop-style discussion, feedback, and community building.

Submission and Review

The submission site and SAML 2027 track details will be announced once they are available. Submissions must follow the IEEE conference proceedings format and use a single-anonymous review process, where author identities are visible to reviewers. Each submission will receive at least three reviews.

All submissions must be original work and must not have been previously published or be under consideration elsewhere. Artifacts, if any, can be shared through repositories such as Zenodo or Figshare and linked from the paper.

Proceedings

All accepted papers will be published in the ICSA Companion proceedings and appear in the IEEE Xplore Digital Library. At least one author of each accepted paper must register for the conference and participate in the workshop to present the paper in person. Papers that are not presented will be excluded from the proceedings.

Submission Deadlines

All deadlines are 11:59 pm UTC -12h (“Anywhere on Earth”).

Abstract submission Dec 16, 2026
Paper submission Dec 20, 2026
Notification of acceptance Jan 19, 2027
Camera-ready submission Jan 29, 2027
Workshop date Either Mar 8 or Mar 9, 2027 (TBA)

Registered Reports & Expected Timeline

A registered report (RR) focuses on the significance of the research question and the quality of the planned study design before the study is executed. For SAML, an RR can be understood as a full research paper without the Results and Discussion sections (stage-1 publication).

RRs are intended to support early feedback on the study design, reduce risks such as outcome-driven reporting or bias against negative results, and provide a structured path toward a full journal paper after study execution (stage-2 publication). They are especially suitable for confirmatory and quantitative studies, but exploratory or qualitative studies can also be submitted when the study design is clear and reviewable.

For a more detailed introduction to registered reports in software engineering, see Ernst and Baldassarre (2023). For details on how ACM TOSEM handles RRs, please refer to their website.

RRs at a Glance

  • Stage-1 submission: detailed study design (full paper without Results and Discussion sections)
  • Page limit: 6 pages + 2 pages for references
  • SAML outcome: reject, accept with revisions, or accept as is
  • Accepted stage-1 submissions will appear in the ICSA 2027 Companion proceedings
  • Post-workshop path: revised stage-1 submission to the ACM TOSEM Registered Report Track, followed by study execution and stage-2 submission

The detailed SAML 2027 RR process is as follows:

  1. SAML solicits RR stage-1 submissions, i.e., the study design descriptions (deadline: see above).
  2. SAML reviewers that agree to review RR submissions also agree to review the stage-2 submissions at TOSEM later on, i.e., the full papers written after study execution.
  3. Each RR stage-1 submission at SAML receives one of three outcomes: reject, accept with revisions, or accept as is (notification: see above).
  4. Accepted RR stage-1 submissions are published in the ICSA Companion proceedings (camera-ready deadline: see above).
  5. At the SAML workshop, accepted RR stage-1 submissions are presented and discussed, with the goal of further improving the planned study.
  6. Authors revise the RR stage-1 paper after the workshop and submit it to the ACM TOSEM RR Track. (deadline: ~1 month after the workshop).
  7. Reviewer contact details for each accepted RR submission are forwarded to TOSEM.
  8. TOSEM invites the reviewers. If no changes were requested, they greenlight the submission; otherwise, reviewers check the updates. The stage-1 submission is not published again at TOSEM.
  9. Once the RR stage-1 submission is accepted at TOSEM, it forms an agreement: if the authors follow the approved study design and reasonably explain any deviations, the final paper can be accepted regardless of whether the results are positive, negative, or inconclusive.
  10. Authors execute the study and extend the stage-1 submission with the results and discussion into a true full research paper (stage-2 submission).
  11. Authors submit the stage-2 submission to TOSEM. The same reviewers are invited. It is treated internally as a major revision (deadline: ~6 months after the stage-1 RR acceptance).
  12. Stage-2 reviews focus on whether the approved study design was followed, whether deviations are justified, and whether the final paper clearly presents and discusses the results. A rejection will only be used if the authors deviated strongly from their initial study design without reasonable justifications.