The ever-increasing programmability of the network has not only enabled unprecedented innovation in networking but also given rise to a new paradigm called in-network computing. By accelerating specific computations on the path between data sources and sinks, in-network computing has shown clear performance benefits for a variety of distributed systems concerns as well as for high-level application components (e.g., for machine learning and scientific computing). However, in-network computing complicates system design, calling for new abstractions and mechanisms. In this talk, I will focus on a fundamental question: How to make in-network computing more accessible? My answer is NetCL---a unified programming system for in-network computing. In-network computing challenges our traditional wisdom in distributed systems, leaving significant future work related to management flexibility, resource elasticity, and security.