PAgE 2027

AI-powered agents now run in production: they plan, call tools, and act on live services and data. Yet the formal methods foundations required to make these systems safe and reliable remain underdeveloped. Ensuring their correctness and reliability demands a new discipline we term agentic engineering that combines principles from programming languages, formal verification, and neuro-symbolic reasoning to analyze, test, and repair agents at scale. This workshop aims to address key questions surrounding agentic engineering and to foster a community of researchers engaged in this area. There is a clear opportunity across the PL, software engineering, formal methods, and machine learning communities to build agentic systems with precise specifications, verifiable behaviors, and runtime checks that hold up in production.

Call for Papers

PAgE (Principles of Agentic Engineering) aims to bring together the formal methods and AI communities to advance the principles of safe agentic engineering. It will feature peer-reviewed papers and invited talks from experts in the field. We welcome submissions describing research results, artifacts, datasets, case studies, and experience reports. Topics of interest include, but are not limited to:

• Specifications and type-safe interfaces for tool use
• Memory and state management for agents
• Reliability & groundedness in planning and tool use
• Static & dynamic analysis and verification of agentic plans
• Safe integration with software engineering workflows
• Testing and debugging of agentic workflows
• Runtimes, compilers and virtual machines for agentic programs
• Evaluation methods and benchmarks for trustworthy agents and agentic software, including safety, reliability, and cost/latency tradeoffs
• Safety, security, and privacy of multi-agent systems
• Neuro-symbolic and formal methods approaches for agent reasoning and control

We also welcome submissions that explore the use of AI agents in the research and development process itself, including agent-assisted ideation, implementation, experimentation, artifact generation, and writing. Such submissions should clearly describe the workflow used, the extent of human oversight, and any lessons learned regarding reliability, reproducibility, and safety.

We encourage submissions describing work in progress, position papers, experience reports, negative results (with detailed analysis and hypotheses for future improvements), demos, artifact or dataset descriptions, and talk proposals.

Accepted submissions will be presented at the workshop but will not appear in the ACM Digital Library. You are free to submit work for presentation that is or will be published elsewhere.

Please ensure that all submissions are anonymized for review. Submissions may be up to 6 pages, excluding bibliography, and should use the standard SIGPLAN conference two-column format.