Process mapping exposes AI queue bottlenecks

Vas argues that enterprise AI pilots stall when they automate individual tasks inside a broken process. His alternative starts at the department level: map the real workflow from systems of record and operator interviews, then assign each step to deterministic software, an agent, or a human checkpoint with evidence prepared. The central claim is that the largest gains come from removing queues, handoffs, and rework—not merely making a task faster. Why it matters: Buying licenses and training people is not a transformation plan. Before funding an agent build, identify where work waits, changes hands, or needs judgment; those boundaries determine whether AI can improve the whole process rather than produce a faster local task.

Try this: Measure one cross-functional workflow this quarter. Baseline its elapsed time, touch time, handoffs, exceptions, rework, and each queue; then label every step deterministic, agent-assisted, or human-approved. Prioritize a bottleneck created by a queue or handoff, not simply the task with the highest volume.