As quantum computing advances toward fault tolerance, the abstractions that once neatly separated applications, compilers, and hardware are beginning to fray. But this is not a failure—it’s a feature. In this talk, I argue that the evolution of better quantum software depends on a continual cycle of breaking and rebuilding abstractions—challenging assumptions when they no longer serve, and crafting new ones to meet the realities of emerging quantum systems. I’ll illustrate this principle with three examples that operate across different layers of the quantum stack.

At the application level, HATT [HPCA 2025] introduces a Hamiltonian-aware fermion-to-qubit mapping breaking traditional fixed mappings to reduce simulation overhead. On the other hand, MarqSim [PLDI2025] introduces a new IR for randomized compilation of Hamiltonian terms. At the compiler level, we proposed a pulse-level compilation scheme that challenges the existing quantum instruction set architecture (ISA) through a series of work [ASPLOS 2019, MICRO 2019, MICRO 2020, HPCA 2023]. Abstractions for supporting these works are adapted in the newest 3.0 version of the OpenQASM language. And at the hardware-adaptive execution level, Surf-Deforme r[MICRO 2024] introduces a low-level deformation instruction set that reconfigures surface code patches in real time to mitigate dynamic defects, redefining how logical qubits are managed under fault-tolerant constraints.

Together, these efforts reflect a broader shift in quantum software: abstraction is no longer a fixed boundary—it’s a tool to be shaped, adapted, and sometimes broken. By embracing this cycle, we can design software systems that are not only more performant and resilient, but also more aligned with the true complexity of quantum computing.