Measuring to Coordinate: Counterfactual Human-Baseline Workload Accounting for Communication and Coordination in Cross-Border Human–AI Teams

When a single founder directs more than twenty semi-autonomous project lines — most of them AI agents, the rest human collaborators — across three time zones, the organisation's work becomes invisible in three ways at once: no native metric makes the output of a human lawyer-hour, an AI research run, and an overnight monitoring watch commensurable; progress is self-reported and therefore unauditable; and coordination depends on the founder's memory of who owes what by when. This paper reports the design and early operation of a management system built to solve all three problems with one instrument. Its core is a counterfactual human baseline: every completed work item, whether produced by a person, an AI agent, or both, is converted into the hours a skilled human would have needed without generative AI, using type-specific benchmark ranges. Progress is never reported; it is measured, as the difference between daily snapshots of milestone-weighted project states. Capacity baselines recalibrate nightly on a rolling seven-window mean, and saturation bands convert measurement directly into next-day dispatch orders, which are written off the following evening in a closed plan–do–check–act loop. In the first audited 24-hour window the system recorded 578.5 counterfactual person-hours — roughly 72 full-time-equivalent staff — produced by an organisation with one principal. We argue the system's larger contribution is coordinative rather than metrological: a shared unit, a shared daily rhythm, and a shared ledger give a cross-border, asynchronous, human–AI team the common ground that co-location and hierarchy normally supply. Patent pending: a United States provisional patent application (No. 64/189,190) covering aspects of the system described in this paper was filed on 1 November 2026.