The context layer has become one of the most talked-about topics in the data and analytics worlds over the past six months or so. But that doesn’t mean that it’s a well-understood concept.
Given its importance to agentic AI projects, it’s critical to have an agreed-upon, working definition. Here, I’ll offer a working definition of what the context layer is, compare it to the semantic layer, and outline why it matters to agentic AI.
Essential Benefits of the Context Layer:
- Adds additional dimensions like time to concepts
- AI agents require context to make accurate decisions
- Without it, the risk of AI agents taking adverse actions skyrockets
The context layer: a definition
Since the context layer is interdependent with the semantic layer, let’s first re-establish what the semantic layer is. (A quick note—at Factor, we’ve historically used the term “information layer,” which is slightly different but loosely correlates to the term “semantic layer.”)
The semantic layer contains your core concepts, vocabulary, terminology, and essentially the conceptual framework of your organization.
Product taxonomies, for example, are a key component of the semantic layer. This is where you’ll find answers to the questions “What is a product?” and “What are the products?”
If you’re a marketing organization, a key concept in your semantic layer would be defining a campaign, the lifecycle of a campaign, the different facets of a campaign, etc. The semantic layer is what creates an organization’s ground truth.
The context layer takes the concepts in the semantic layer and adds additional dimensions. To put it simply, the context layer goes beyond answering the question “What is a campaign?” to “What is a campaign at our organization?” or “What is a campaign at this moment in time?” Data lineage, governance, and access control are also fully integrated, creating a single, holistic surface.
A brief clarification: there is (and likely will continue to be) some debate among practitioners about whether the “context layer” is a new construct, or simply a rebranding of existing semantic and metadata capabilities. In our experience working with enterprises on semantic layer projects for more than a decade, many concepts that are now recognized as the context layer have always been part of the conversation. As with everything in the semantic realm, the vocabulary will continue to evolve.
What “context” looks like in practice
Let’s say you have a concept that is fundamental to your semantic layer—we’ll keep using “what is a campaign?” As mentioned earlier, one way to think about the context layer is that it takes those fundamental concepts and adds dimensions, like time, to them.
That means that in the context layer, you (or an agent) might want to understand questions like “What is the process that a campaign goes through over time?” That process will be specific to your organization or department.
Throughout that process, there’s lots of both structured and unstructured data that gets generated and lots of concepts that need to be tracked. How does the data flow in, through, and out? What is the output? What are the events that take place throughout the lifecycle? These are the kinds of data points, specific and unique to your organization, that can be managed in a context layer.
Why does agentic AI need the context layer?
Without context, the risk of agentic AI skyrockets. Remember, AI agents are designed to take actions. And make decisions. If a model is going to make a decision, it obviously needs to understand the decision criteria, but that’s not all. It also needs to know the decisions that came before. It needs to understand the environment the decision is being made within, and it sometimes needs to adjudicate between different views of concepts that are part of the semantic layer. A context layer allows for the audit trails and precedent records that an AI agent needs in order to act reliably.
If you think of your semantic layer as, in effect, your conceptual registry, where a canonical notion of things like “campaign” are stored, then the context layer is where all of the different implementations, aspects and facets of the “campaign” concept can be reconciled. This overlay (which often consists of “active metadata,” and which can itself be stored in a “dynamic” graph) is where the agent can ingest the context necessary to execute dependably.
Without it, the agent simply does not have the context needed to make reliable, effective decisions, so you run the very real risk of the agent taking actions that have adverse consequences.
This is why so many enterprises are investing in their information architecture now–the risk of neglecting your semantic foundations, of which the context layer is one part, is simply too great.
Of course, knowing that it’s time to invest in your context and semantic layers and knowing where and how to begin are two different things. If you’re struggling with where and how to start, we have resources:
- Learn why a stable foundation is necessary for any AI tool
- Hear from two of our IAs and our co-founder Gary Carlson about milestones and pitfalls of building your AI foundations
- Listen to our co-founder Bram Wessel and Profisee CDO Malcolm Hawker discuss what a successful AI initiative looks like
- Explore our AI Activator program to jumpstart your work