# AIOA

AIOA places a large-language-model coding agent in a keep-or-discard loop that rewrites the numerical algorithms of the atomistic simulation library ASE, with verifiers the agent cannot alter guarding each keep. Across three independent runs, six numerical kernels ran 22–24 times faster on average in Python with NumPy and SciPy while returning unchanged results; on blind held-out problems, optimizers from the best configuration needed 3.1–3.4 times fewer force evaluations than L-BFGS for structure relaxation and 4.6–17 times fewer than standard CI-NEB for reaction paths.

AIOA places a large-language-model coding agent in a keep-or-discard loop that rewrites the numerical algorithms of the atomistic simulation library ASE, with verifiers the agent cannot alter guarding each keep. Across three independent runs, six numerical kernels ran 22–24 times faster on average in Python with NumPy and SciPy while returning unchanged results; on blind held-out problems, optimizers from the best configuration needed 3.1–3.4 times fewer force evaluations than L-BFGS for structure relaxation and 4.6–17 times fewer than standard CI-NEB for reaction paths.