LLVM's intrinsic infrastructure in TableGen has been evolving rapidly to keep pace with the growing number of target-specific intrinsics, particularly in GPU and ML accelerator targets. In the first part of the talk, we present a set of enhancements that make intrinsic definitions more expressive, extensible, type-safe, and precisely validated: pretty-printing of immediate arguments, default argument support, explicit overload type constraints, and disjoint range sets. In the second part, we cover additional improvements to the intrinsic verification and encoding infrastructure: IIT encoding refactoring, and better diagnostics for vararg and overload signatures. Together, these improvements enhance the usability of the infrastructure - practical know-how for LLVM developers looking to leverage these additions.