Within 28 days in late 2025, the internet buckled three separate times under outages at different large-scale cloud providers. These failures underscore a hard truth: While cloud infrastructure delivers enormous benefits, its ubiquity also creates large, systemic points of failure. CyberCube estimated that the first AWS outage alone caused nearly $581 million in losses, affecting everything from major streaming platforms to the now-infamous Eight Sleep smart bed debacle. Few would argue for a return to the flash-drive, pre-cloud era. However, many engineers have begun questioning our relationship with these massive cloud structures. One viable reconfiguration, known as 'local-first' and based on a 2019 research paper of the same name, reframes the cloud as an on-demand task runner and simple synchronisation layer between devices rather than the backbone of every user interaction. While local-first isn't a silver bullet – Aaron Boodman of Rocicorp famously describes the space as a 'big, beautiful, hot mess' – early experimentation has produced encouraging results. Several production-ready tools, including the highly anticipated TanStack DB, have recently entered early adoption, suggesting that local-first may be on the cusp of going mainstream. These tools have shown the potential not only to improve reliability, but also to simplify entire technology stacks by sidestepping cloud-centric problems such as caching, authorisation flows, brittle error logic, race conditions and tightly coupled state management. Ultimately, this talk argues that local-first isn't about rejecting the cloud, but renegotiating its role. By moving core user experiences back onto the device and treating the cloud as a partner rather than a dependency, we can build applications that are faster, more resilient and more humane. As outages become less exceptional and more expected, local-first may offer one of the most practical paths towards a healthier, more durable web.