At Photon, we are developing the world’s most complete iMessage API (with RCS and SMS fallback) for agents: infrastructure that lets Agents reliably communicate through one of the most native messaging interfaces people already use. While building it, we run Mac-backed systems continuously and care a lot about making the platform stable enough for enterprise-grade workloads. During that work, we ran into a macOS networking issue that showed up after about 49.7 days of uptime: new TCP connections stopped working, while ping and existing connections still looked normal. We reproduced the behavior on two machines and traced it to how XNU handles a 32-bit TCP timestamp counter around wraparound. This post walks through what we saw, how we reproduced it, and the kernel code path behind it.