Agentic engineering is an emerging discipline focused on the design, development, and operation of systems that exhibit goal-directed autonomy. Foundation models (FMs), such as large language models (LLMs), have been accelerating progress in this area across academia and industry.

Agentic systems often involve multiple interacting agents, humans, and tools, requiring rigorous system-level engineering to ensure critical qualities like robustness, safety, and observability. A key design challenge in agentic engineering is the growing capability of FMs/LLMs. Developers must decide whether to rely on the FM/LLM or external tools/systems for the same functionality. These decisions can be made at various stages depending on the problem and context: during design time, development time, or even at runtime from a software engineering perspective, and at pre-training time, post-training time, test/inference time, and post-inference time from an AI perspective. Highly autonomous agentic systems also require continuous monitoring, evaluation, observability, intervention, and oversight after deployment—an emerging discipline referred to as AgentOps.

While workshops on agentic or multi-agent systems have appeared at AI-focused venues, they typically emphasize theoretical modeling, multi-agent learning, or coordination protocols. In contrast, this workshop is situated within the software engineering community and addresses concrete engineering methods, design trade-offs, and operational practices needed to develop and maintain agentic systems built on foundation models.

We also recognize that agentic engineering builds on foundational work from the agent-oriented software engineering (AOSE) community. However, the emergence of foundation models introduces new challenges around autonomy, tool integration, prompt-driven behavior, and post-deployment adaptation. This workshop seeks to update and re-contextualize those principles to address the design and assurance of modern agentic systems, particularly those grounded in large-scale pretrained models.

This workshop will provide a forum for exploring engineering methods, techniques, and tools for agentic systems in general and agentic systems for software engineering in particular. It will bring together researchers and practitioners to share insights, innovations, and real-world experiences in the design, development, and operation of agentic systems.

Topics of interest include, but are not limited to:

- Requirements engineering for agentic systems

- Architectural design for agentic systems

- Verification, validation, and testing of agentic systems

- AgentOps – DevOps for agentic systems

- Development processes and lifecycle management for agentic systems

- Evaluation methodologies, tools, and benchmarks for agentic systems

- Responsible AI and AI safety of agentic systems

- Agentic systems for software engineering, including requirements, design, coding, testing, deployment, and operations

- Human-agent interaction, collaboration, and oversight

- Risk and impact assessment (e.g. economic/social impact)

- Real-world case studies and practical experiences in different domains

All submissions must be in English, in PDF format, and must not exceed the page limits (including references and appendices) listed above. The workshop follows a single-anonymous review process. Submissions should be made via HotCRP (https://icse2026-agent.hotcrp.com/). Note this year the official “ACM Primary Article Template” should be used for submissions.