The Fix
Autonomous units remove coordination as governance but do not remove the need for shared understanding. The medium changes: coherence is maintained by AI systems that read versioned artefacts continuously and produce synthesis on demand.
The first synthesis run produces seventy-three discrepancies. The unit lead reads the output on a Tuesday morning, scrolling through a list that is longer than she expected and noisier than she hoped. Most of the flags are definitional: states the process definition names that the code handles differently, transitions the code implements that the definition omits. She triages for forty minutes before she finds the first real one: the settlement calculation applies a discount threshold that the process definition does not mention, because the threshold was added eleven months ago by an engineer who updated the code but not the definition. The settlement has been silently discounting claims above a certain value. Nobody noticed because the output looked plausible.
The unit lead marks the flag, opens the process definition, and records the threshold as an undocumented deviation requiring review. It is the first time the artefact has forced a silent discrepancy into the open rather than quietly inheriting it. Autonomous business units remove the need for coordination as governance, but not the need for shared understanding. What changes is the medium: coherence is maintained by AI systems that read versioned artefacts continuously and produce synthesis on demand, rather than by heroic individuals or social negotiation. Three things make this work: the artefacts, the synthesis layer that reads them, and the controls that connect the two.
The artefacts sit within a structural hierarchy, each level deriving from the one above it. Strategy produces a small set of falsifiable commitments, which the process catalogue maps to the organisation's value-creating processes and assigns to owners. Autonomous units implement the processes they own, each within a bounded context that defines what data, services, and decisions belong to it; operational data lives inside the boundary, serving the process rather than the reporting layer, and contracts govern what crosses it, versioned machine-readable interfaces through which units interact without coupling. The analytical domain sits downstream, consuming what units publish through data contracts, never dictating what they store. AI reconciliation reads the full chain continuously, from strategic commitment through process definition through system telemetry, and reports where it breaks. When any level is implicit, everything below it degrades into social negotiation, and the AI reader has nothing to reconcile against.
The distance from the inherited model is visible in the contrast:
0.45em
Old model. Strategy (narrative) $$ Portfolio governance $$ Shared systems $$ Shared data $$ Coordination $$ Narrative reporting.
New model. Falsifiable strategy $$ Process catalogue $$ Autonomous business units $$ Bounded contexts $$ Operational data $$ Contracts $$ Analytical domain $$ AI reconciliation.
Four categories of artefact form the substrate of coherence. Each must be precise enough to be read by a machine, versioned, and owned.
...
Continue reading in the interactive reader
Read this chapterSee also: Full contents · Preview chapters · Illusions in the Boardroom