The ecosystem lens
This is the map we start from in almost every workshop and talk. It is the answer to a question that comes up constantly: if the model is a commodity, what actually makes an enterprise AI system work?
Reading the map
Each layer only makes sense once the one beneath it is in place. Most failed AI projects skip straight to the top.
Microsoft Foundry, the model catalogue, and the platform services around them - search, document intelligence, content safety, evaluation and observability. Security, governance and observability sit alongside rather than on top: Agent 365, Entra and Agent ID, Defender, Purview, Azure Monitor.
Every enterprise AI system starts with data. Work data sources, enterprise data sources, and the tools, APIs and services reached through MCP. This is where most of the real engineering effort actually goes.
Cross-cutting semantic and reasoning intelligence across the Microsoft cloud. Work IQ carries human and organisational context, Fabric IQ enterprise data understanding, Web IQ fresh real-world intelligence, and Foundry IQ enterprise knowledge and retrieval. Four kinds of context, not four products.
Low-code agents in Copilot Studio, data intelligence agents over Fabric and Databricks, pro-code agents built and hosted on the Microsoft Agent Framework, Foundry Agent Service and Container Apps - and federated agents running on other platforms entirely. Agent 365 observes, governs and secures the estate across all of them.
Where AI becomes visible to people: Copilot and M365, Copilot Studio channels, data-native experiences, pro-code custom interfaces, and the enterprise SaaS systems your organisation already runs on.
The point
It’s an intelligence platform that connects data, models and work. That is why so much of what decides success sits below the waterline - use case selection, architecture, security, data and governance - and why so little of it is about the model itself.
See what we teach →