A new name for the layer you already own

Boris Evelson and Indranil Bandyopadhyay proposed a category on 20 August. A context layer, in their definition, is the next evolution of semantic layers and knowledge graphs — the foundation for neurosymbolic context engineering and agentic applications.

What it is made of

• Business semantics and governance, from the semantic layer.
• Ontological modelling, from the knowledge graph.
• Enterprise knowledge across data, metadata, business concepts, policies and processes.
• Runtime context — events, decisions, actions and outcomes.
• Graph-based ontologies with linked contextual relationships.

The phrase the authors settle on is a living model of the enterprise that enables AI reasoning, automation and decision intelligence.

The one genuinely new ingredient

Four of those five items are things a mature data organisation has already bought, in some form, more than once. The addition is runtime context: not only what the business means by a term, but what happened, what was decided, what was done and what resulted. A semantic layer describes the world as modelled. A context layer is being asked to describe the world as it went — and that is a data collection problem before it is a modelling one.

How the market is being carved

Forrester sees two subcategories forming: capabilities embedded in business applications, from application vendors and hyperscalers, and general-purpose ones from graph database, catalog and lakehouse vendors. The first arrives with context already populated and a boundary around it; the second arrives empty and unbounded. Which is the better buy depends entirely on whether the decisions being automated stay inside one application.

A Landscape report is scheduled for the end of Q4 2026, with a Wave to follow. The blog names no adoption figures and makes no market forecast, which is the honest position for a category this early — and a reason to treat the term as a procurement question for next year rather than this quarter.