Researchers have proposed StagedWorkspace, a versioned workspace for knowledge-work agents. This system binds parsed records and review diffs to content hashes of native files as they change. In experiments on OfficeQA Pro and APEX-Agents, dual parsed/native access improved model performance by up to 12.1 points and 9.2 points respectively. The authors suggest that workspace state is an experimental variable in knowledge-work agents and propose benchmarks that score evidence, staged edits, and submitted artifacts as explicit state transitions. --- Why it matters: This matters because it highlights the importance of considering workspace state when developing knowledge-work agents. By explicitly tracking changes to a workspace, these systems can improve performance and accuracy in tasks such as code review and document editing. Source: https://arxiv.org/abs/2608.18050