In case you missed it as I did, just before the 2025 holiday season, Jaya Gupta (a Partner) and Ashu Garg (General Partner) of Foundation Capital declared a new trillion dollar market, one they’re calling “Context Graphs”. 

Since then, Jessica Talisman noted a high number of similarities between the language in her November 2025 Process Knowledge Management Parts I and II on Substack and the December 22, 2025 Gupta and Garg article. I refer you to her January 27 LinkedIn post to review her assertions.

The term “context graph” has been around awhile, but Gupta and Garg gave the term a more specific definition related to systems of agents. I’ll explore this point more later. 

Googler search frequency on “context graph”

Using agents to build and harness the power of contextually-specific graphs is not a new thing. I remember meeting Dave Duggal of EnterpriseWeb (EW) at a Dataversity semantic technologies event back in the 2010s. In 2015, EW helped HP use agents working in HP’s own databases to compile monthly reports, reducing the time taken to less than a second. EnterpriseWeb ended up doing a series of proofs of concept for the firm I was with then. EW’s approach as far as I’m aware has always been human-first, human-in-the-loop, but agent-enabled. 

Since that time, EW has been one of the pioneers behind network function virtualization (NFV) for the telecom industry, abstracting the complexity of networks of switches.  In 2025, EW announced its Telecom Ontology at an IPWC (a wireless networking consortium) event. The ontology is at the heart of EW’s interoperability solution that can run inside Snowflake, one that harnesses the power of Telecom Management Forum (TMF) and other industry standard concepts and metadata.

Yet Another Trillion Dollar Market

Proclaiming a new trillion dollar market is not a new thing. Back in 2008, John Doerr, Chair of Kleiner Perkins, declared green energy a trillion dollar market. Since that time, Doerr has been backing related sustainability efforts, including a $1.1 billion donation to Stanford to build the Stanford Doerr School of Sustainability in 2022. 

Next Move Strategy tracks the global renewable energy market and pegged that market at $856 billion in 2021. A current forecast through 2030 predicts a market size of just over $2 trillion, a compound annual growth rate (CAGR) of just less than ten percent.. So by 2023, green energy may well have become a trillion dollar market. 

I agree that contextual computing will eventually be a big thing, but context by that definition is enormous in scope. As I’ve said many times before, John Launchbury at DARPA back in 2018 declared the Third Wave of AI–adaptive context, also known as contextual computing, the blending of knowledge representation (today’s knowledge graphs) and statistical machine learning methods. The semantics/knowledge graph community at large has literally been focused on creating machine readable context with the help of knowledge graphs and natural language processing (NLP) for decades. 

Context Graph Defined

Gupta and Garg argue that because systems of agents sit in the middle of a workflow, crossing boundaries between operational applications, agent systems can record state at the moment of a commit or decision, creating an audit trail that traditional databases miss.

Core components of a context graph, per Jaya Gupta and Ashu Garg, include the following:

A decision trace captures the specific events, exceptions, overrides, and precedents that led to a decision. This includes tacit information that usually lives in Slack threads, emails, or employees’ heads.

Stitched entities link data across different systems (e.g., connecting a support ticket in Zendesk to a contract in a billing system and a churn warning in Slack) to show a more complete picture than any single application has of the background behind a decision.

Traces over time  form searchable precedent,  a historical record. Such a history conceivably allows AI agents to look back at past exceptions (e.g., “Why did we give a discount to a similar client last quarter?”) and use that context to improve the effectiveness of future actions.

A Context Graph as an Operationalized Knowledge Graph

Most business intelligence is tacit. It lives in people’s heads, emails, and chats rather than in transactional databases. In theory, Context Graphs capture an audit trail of these informal interactions so that AI doesn’t just follow rigid rules, but understands the intent and exceptions that make a business actually run.

In that sense, Andreas Blumauer of Graphwise points out that a context graph is a kind of knowledge graph.  “The consensus across the community ,” he says, “is that a Context Graph is an operationalized Knowledge Graph.”

Blumauer notes that higher level abstractions of a knowledge graph (such as domain specific usages of “Customer”) make operationalized graphs possible:

“The Foundation: You cannot have a Context Graph without the entities defined in a Knowledge Graph. (You can’t trace a decision about a ‘Customer’ if you don’t have a ‘Customer’ node).”

Moreover, knowledge graphs can be either static or dynamic. Blumauer continues:

“The Extension: The Context Graph adds a meta-layer to the Knowledge Graph, often using technical standards like RDF-Star or Named Graphs to attach timestamps, provenance, and confidence scores to the links between nodes.”

Savvy database technology users often harness the power of immutable storage to enable versioning. Temporality is part and parcel of immutability, as knowledge graph platform vendor Fluree points out: “Immutable databases use cryptographic signatures and temporal metadata to record important information about the ledger of changes: when a record was created, updated, and pointers to previous versions.”

Timothy Cook of Axius SDC in a recent interview posted in these pages talks about immutable versions of ontologies. Space, time and event-oriented modeling are all capabilities that semantic knowledge graph developers can take advantage of today.

The bigger picture is semantics graph RAG and various human-agent feedback loops for different purposes throughout the enterprise, not just agents that do workflow tracking and recording. A feedback loop-rich, interoperable environment across supply chains is how the contextual computing market could become a bona-fide trillion dollar market.

For More Information

Alan Morrison, “In Favor of Contrarian AI,” The GraphRAG Curator, November 11, 2025, https://graphrag.info/2025/11/11/in-favor-of-contrarian-ai/.

Andreas Blumauer, “How do Context Graphs and Knowledge Graphs differ from each other?,” LinkedIn, February 1, 2024, https://www.linkedin.com/pulse/how-do-context-graphs-knowledge-differ-from-each-other-blumauer-wkfcf

Huang, Q., Chen, P., Kržmanc, G., Ren, H., Liang, P., Zeng, D., & Leskovec, J. (2023). PRODIGY: Enabling In-context Learning Over Graphs. Advances in Neural Information Processing Systems, 36. https://proceedings.neurips.cc/paper_files/paper/2023/file/34dce0dc3121951dd0399baJ

Jaya Gupta and Ashu Garg, “AI’s Trillion-Dollar Opportunity: Context Graphs,” Foundation Capital (blog), December 22, 2025, https://foundationcapital.com/context-graphs-ais-trillion-dollar-opportunity

Jessica Talisman, “Process Knowledge Management, Part I,” Intentional Arrangement (Substack), October 31, 2024,https://substack.com/@jessicatalisman/p-180668324.

Jessica Talisman, “Process Knowledge Management, Part II,” Intentional Arrangement (Substack), November 7, 2024, https://jessicatalisman.substack.com/p/process-knowledge-management-part.