This talk will describe the reasoning and mechanisms for maintaining an LLVM production toolchain that lives very near the tip-of-tree (ToT). We will discuss our reasoning (we like to incorporate features, security benefits and performance improvements as quickly as possible). We will describe at a high level multiple layers of infrastructure we use to maintain this toolchain, including our strategies for testing, cherrypicking, bad-commit isolation and human/AI processes around the integration process. We believe that the lessons learned, mechanisms developed and arguments for investing in a tip-of-tree toolchain are relevant across the industry.