Meridian's data engineers are building the ingestion-to-model pipeline for the customer-service chatbot and want it optimized specifically for AI workloads rather than reused unmodified from an existing generic reporting-ETL pipeline. What distinguishes an AI-optimized pipeline from a standard business-reporting ETL pipeline in this context?

Constructing data pipelines optimized for AI workloads means the pipeline does more than a standard reporting ETL: it must support feature engineering, preserve and manage labeled training data, handle model-specific input formatting, and often support both batch training and low-latency inference paths. A generic reporting pipeline, built to populate dashboards, typically doesn't handle labels or produce model-ready feature vectors. The claim that nothing distinguishes them and any existing ETL pipeline is equally suited wrongly assumes plug-and-play equivalence, ignoring the additional, AI-specific requirements the outline calls out. The claim that AI pipelines never require any transformation step is an overstatement — transformation steps exist in both pipeline types, so that statement is simply false. The claim that AI-optimized pipelines only differ in the color-coding of their diagrams trivializes a substantive engineering distinction into a cosmetic one. For the exam, remember that reusing an existing BI/reporting pipeline unmodified for an AI initiative is a common project pitfall — the pipeline must be purpose-built or substantially adapted for AI-specific needs like labeling and feature generation.