How "Context Graphs" are Redefining the AI Landscape

Welcome to a new edition of The Board: Distillation Aftershots ^(*).

This is an online post of a weekly newsletter that shares curious and interesting insights and data points distilled from enterprise technology to identify what’s notable.

In this issue, we are going to start – so much to cover – addressing context as the next AI evolutionary hurdle. And no better way to do that than by sharing a few links and ideas on Context Graphs – the “revolutionary” (read marketing) theme that brought context to the forefront inside the AI community.

First, my take.

Context graphs are a marketing aberration created by mixing two different things: context (the ability of computers to understand the right space, time, and need by using ancillary information to content), and graphs (one of the core elements that helps context by visualizing relationships between the elements and the situation; you became aware of it during the social evolution, if not before during knowledge management like me).

While it is a simplistic way to marry the concepts, it robs context of its true value of going beyond graphs (I said it was one of the core elements, together with memory and metadata). Focusing on graphs as the only way to represent context simplifies the concept and – more importantly – creates a “market segment” where vendors and funding partners can play to their strengths.

It helps explain complicated concepts to people who may not have the right mentality. Anyone who worked in Knowledge Management since the 1960s (most of us did it in the 1980s and 1990s) understands the knowledge triumvirate: content, context, and intent. The intricate relationship among the three of them engenders the one-answer idyllic model: every question has a single, correct answer when framed within those three elements. Content is the what, context is the how, and intent is the why (when and who are determined by who’s asking).

GenAI is entirely about the content: it regurgitates content that ingested and indexed. It is also a limitation, as no question can ever be answered properly without context (the metadata that frames the use of the information requested). Whether a user asks about a loan because they are about to buy a house, a car, or want to throw a crazy party sets the context (and intent, that’s for much later) to determine which loan and process would work better. Agents need that context if we want them to properly automate transactions (the ultimate goal for Agentic AI).

Thus, a simpler way to think of context is to say: we can build a graph of how the agent must make decisions to determine context, which will simplify the decision tree it needs to use (and allow us to create a product to answer). Think about context graphs as “journey management” in the CX world. While it is impossible to construct a single experience for everything, we can pretend we can do so by mapping the myriad elements of the journey, and a journey to get there. True, we are likely to miss certain elements and characteristics, but we are reducing the number of potential answers, thereby increasing the certainty of the right, unique answer. Context graphs are the journey management of the agentic AI world.

I oppose journey management for oversimplifying a very complicated process, yet it still exists and is sold as a product. Thusly, enter context graphs. Will look into better ideas for context management (context engineering is making the rounds, that’s next) that don’t reduce a complex concept to a simpler product-to-be-sold.

Here are some reading resources (this is a very early concept; most of the content is still in LinkedIn discussions): 

1. Let’s start with the article that started it all, the introduction to her post on LinkedIn, and more posts from the same author to further understand where this is coming from. There is really good material in here. 
2. Dharmesh Shah, HubSpot’s CTO and an ardent fan of agents, wrote a very detailed explanation of what he sees as the reason for Journey Graphs, worth the read. 
3. Another of my fellow Graphs enthusiasts has a detailed explanation of the good and bad of Context Graphs (and a great comment section with an excellent discussion) 
4. A decent argument in favor of Context Graphs in the world of customer service, sales, and marketing (with a little bias towards their product). 
5. For balance, a good discussion on why context is not a graph, with a link to a great article with more detail.

What’s your take? We are fostering a community of executives who want to discuss these issues in depth. This newsletter is but a part of it. We welcome your feedback and look forward to engaging in these conversations.

If you are interested in exploring the full report, discussing the Board’s offering further, or have any additional questions, please contact me at [email protected], and I will be happy to connect with you.  Would you to get the newsletter? Subscribe.