Deep research agents tackle complex questions by looping through search, reading, and reasoning. The conventional design treats each search as a discrete action interleaved with reasoning. OpenSaC, a new Python library described in an arXiv preprint (2610.25779), proposes a different framing: "Search as Code."
By making search a programmable, code-like component, the library aims to move away from ad-hoc search-and-read steps. This could make research pipelines more modular and easier to reproduce, since search logic becomes explicit and reusable rather than embedded in an agent's step-by-step decisions.
The abstract is short and does not provide implementation details or evaluation results, so it is too early to judge the library's practical impact. Still, the conceptual shift is notable: treating search as a first-class programming construct may help researchers build more transparent and maintainable deep-research systems.