Workshop: AgenticAI4HPC’26: The First International Workshop on Agentic AI for HPC
Authors: Yuzhong Shen and Masha Sosonkina (Old Dominion University) and Peng Xu and Mark Gordon (Iowa State University)
Abstract: Modernizing legacy Fortran is a problem of volume: the transformations are individually routine, but the codebases are enormous, and the work goes undone. We propose an agentic workflow operating at production scale. Three prompt-specialized agent roles operate under a version-controlled specification the agents themselves authored, while humans hold a small number of gates. Safety is anchored to an exact verification oracle inherited from the domain.
We apply the workflow to the two-electron-integral routines of GAMESS (General Atomic and Molecular Electronic Structure System), converting twelve source files (56,448 lines, 225 subroutines) from fixed-form Fortran 77 to free-form Fortran 2008. The agents ran as Claude Code roles in isolated worktrees. Because GAMESS's user community treats its printed test energies as canonical, we adopt bit-for-bit reproduction as the merge criterion. All twelve files pass the 51 standard tests with zero chemistry-relevant differences and the Jenkins continuous-integration test suite.
Back to AgenticAI4HPC’26: The First International Workshop on Agentic AI for HPC Archive Listing
Back to Full Workshop Archive Listing