Building Agentic-AI Frameworks at Scale – April 2026
Overcoming the roadblocks to system wide agentic-AI implementation
McKinsey’s latest report on Agentic AI focuses on “strong data” – consistently formatted, high quality, data made continuously available to an agentic-AI framework. McKinsey notes:
“Nearly ⅔rds of enterprises have experimented with agentic-AI but fewer than 10% have scaled to deliver tangible value”
To address this, McKinsey argues that an “agentic-AI data architecture” needs to be implemented to break down data silo’s and improve data visibility. This requires the introduction of a semantic layer between the data and the AI applications. This layer comprises a data dictionary, an ontology and a knowledge graph. The dictionary defines the data terms, the ontology converts these into machine-readable relationships, and the knowledge framework maps each terms to its context using the unique ID attached to each data term. McKinsey concludes:
“Organisations need to engage in continuous, real-time data quality monitoring. This process must be supported by automated validation, anomaly detection, and enrichment pipelines that prevent issues propagating across workflows.”
This requires continuous observability over the source and destination of data, together with a precise understanding of how data is manipulated as it migrates across sets of co-operating services.
This is achieved by introducing a Topology Control Plane that captures every message initiation (send) and termination (receive) event, and maps these interactions into a graphical formalism to deliver a high-fidelity representation of the runtime telemetry. This continuously updating system representation provides continuous observability of the source and destination of data, and a precise understanding of how data is transformed as it moves across co-operating services. Only by establishing continuous observability of the runtime telemetry can assurance be given that consistently formatted, high quality data is being continuously provisioned to the implementation’s agentic-AI framework.
A full copy of the McKinsey article Building the Foundations for Agentic AI at Scale may be found here: