AI advice can help groups coordinate, but it may also allocate the costs of coordination unequally. We examine this trade-off in a repeated three-player threshold public-good game, where efficiency requires the two lowest-cost players to contribute. An LLM advisor provides either private, personalized advice or public, group-level advice. Among LLM agents, group advice almost eliminates coordination failure, but it does so by concentrating contribution costs on particular players. Human participants respond differently. When exposed to the same advice, they often sacrifice efficiency for equality: they contribute less, provide the public good less frequently, and reject costly contribution recommendations, yet achieve a more equal distribution of net payoffs. These findings suggest that AI advice can resolve coordination problems, but it does not eliminate the distributional conflict generated by efficiency-enhancing recommendations.