AI is attracting significant attention and experimentation, but its adoption across public institutions remains uneven. Turning technological possibility into public value means balancing impact and efficiency with transparency, impartiality, accountability and sovereignty. That requires more than strategic evaluation: it depends on the operational choices made by staff with expertise specific to their institutions. This session explores how practical guardrails, alongside formal regulation, can give public servants room to experiment while ensuring AI serves the institutions and citizens it is meant to support.