So close, yet so far: can AI succeed where health systems are failing?

Abstract Artificial intelligence (AI) is increasingly positioned as a transformative tool in health, including in sexual, reproductive, and maternal health (SRMH). Yet this momentum coexists with a more uncomfortable reality: many health systems continue to face foundational challenges, weak infrastructure, fragmented services, workforce shortages, poor data quality, and persistent inequities that are not technological in origin. This article argues that AI's potential contribution to SRMH will be shaped more by the conditions under which it is introduced than by its technical sophistication. Four interconnected arguments are developed. First, rapid adoption should not be mistaken for demonstrated value; the evidence base for scalable, equitable impact remains limited. Second, effective AI integration requires starting from health system bottlenecks rather than technological possibilities, since many SRMH barriers are organizational, political, and cultural before they are computational. Third, institutional readiness, governance, absorptive capacity, and implementation infrastructure are central determinants of whether AI improves or complicates care. Fourth, without deliberate policy decisions, AI is likely to benefit populations already better served, amplifying existing inequities. A pragmatic agenda for AI in SRMH must therefore begin with what health systems need, not what technology can offer.