Novolex's move from RPA scripts to AI agents exposes the real constraint: most RPA delivers marginal cost cuts while agents unlock process redesign. Automation Anywhere's case shows measurable throughput lifts in packaging workflows once decision logic moved beyond rules. The ROI hinges on shrinking exception handling, not bot count. Yet without clear baselines on cycle time and error rates, these pilots often inflate hype over hard returns. What integration friction has slowed your own agent rollout most? Full video link https://youtu.be/R_VhkPh48Jg