This presentation gives a high-level overview of ongoing research that is aimed at using Ai/ML in E3SM. There are many projects funded by BER, lab LDRDs, and other funding that are looking to exploit AI/ML to improve E3SM. The improvements can range from process emulation, bias correctors, calibration, and diagnostics. Some exemplars will be shown, as well as progress towards a community of practice. The talk will include a discussion (not answers or results, just a discussion) of what software infrastructure could be supported that would (1) lower the barrier to further integration of ML emulators into E3SM and (2) provide a template for how to achieve performance-portable implementation for each of Fortran and C++.