The site describes a research program spanning environments, evaluation, training, and harnesses. Its thesis is: Office Intelligence treats continuing work environments—with evidence, review, and retained context—as the research object for knowledge-work agents.
The mechanism is: A task-time workspace combines authorized files, tools, organizational state, and task context; deliverables are checked against task-specific criteria and evidence links, while execution traces and working memory can inform later tasks. The site also gives an OfficeTown example involving a reconciliation role and a reviewable workbook. The why_now is: Long-running work makes continuity, evidence, and review part of the agent interface rather than optional prompt context. The first implication is: Evaluation can focus on whether an agent produces an inspectable artifact under task-specific constraints, not merely whether it answers a prompt.
The evidence boundary is: The public site describes a research program, environments, and examples; it does not establish validated performance in real organizations or prove that context reliably transfers across tasks. The first watch condition is: Watch for published evaluations showing whether retained work traces improve later-task outcomes without weakening evidence quality or reviewability. If those results appear, the project’s framing can be assessed as more than a research design. No prior New Runtime record is supplied, so this starts a distinct analytical record on continuing work environments as an agent research and evaluation object.