The authors propose a system‑level view of work agents as compositions of foundation models and runtime harnesses that operate within stateful workspaces to produce verifiable outcomes. They define two orthogonal conditions — work persistence and work openness — to describe the environment in which tasks are performed, and use these to partition digital work into five categories based on task scope and organizational structure. Leveraging this taxonomy, they enumerate five core capability families necessary for dependable execution, and they investigate three optimization strategies: independent model optimization, independent harness optimization, and joint model‑harness optimization. The paper further surveys existing evaluation protocols, emphasizing outcome‑centric metrics, and outlines future directions including sustained delegation, adaptation to open‑world settings, and the feedback loop between execution performance and capability refinement. This integrated framework allows systematic analysis of agent reliability and facilitates joint optimization of model parameters and harness policies, thereby improving task success and verifiability.