--- title: "Context Graphs Are a Trillion-Dollar Opportunity. But Who Actually Captures It?" url: "https://atlan.com/context-and-chaos/issue/context-graphs-are-a-trillion-dollar-opportunity-but-who-captures-it/" description: "Jaya Gupta's thesis is right about context graphs, and wrong about who wins. In a world of heterogeneity, the integrator always wins, not the application." keywords: "Context Graphs, AI Infrastructure, Enterprise Data, Market Analysis" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ A Context & Chaos deep dive by Prukalpa Sankar (Co-founder & Co-CEO, Atlan), published January 13, 2026 (11 min read). It responds to the viral context-graph debate: Jaya Gupta's thesis is right about context graphs and wrong about who wins. In a world of heterogeneity, the integrator always wins, not the application. ## The debate - Jamin Ball's ["Long Live Systems of Record"](https://cloudedjudgement.substack.com/p/clouded-judgement-121225-long-live) argued AI agents won't kill systems of record (warehouses, CRMs, ERPs, HRISes); "truth" will live there with a semantic layer telling agents how to use it. - Jaya Gupta and Ashu Garg replied in ["AI's trillion-dollar opportunity: Context graphs"](https://foundationcapital.com/context-graphs-ais-trillion-dollar-opportunity/): semantic layers miss **decision traces**, the "why" behind past decisions, which together form a **context graph**, the next trillion-dollar opportunity. They argue agents in the "execution path" will own it, so vertical agent startups own the context graph for their domain (renewals for the sales agent, escalations for the support agent). - Follow-on discussion covered how to build a context graph, two-layer context architecture, operational context and execution intelligence, and implications for agent reliability, governance, observability, AppSec and financial recovery. The author agrees on the importance and opportunity of context graphs, disagrees on who captures it, because the idea runs into enterprise **heterogeneity**. The post draws on 18 to 24 months of customer conversations, Atlan's product strategy, and her essays on the [enterprise context layer](https://atlan.com/know/closing-the-context-gap/) and the [AI context gap](https://atlan.com/know/ai-value-chasm/). ## The end of one heterogeneity, and the start of another For a decade, data heterogeneity meant point tools orbiting a few closed warehouses, then consolidation. Iceberg and open table formats are ending that era: storage is open, compute fungible, lock-in moving downhill. Heterogeneity is moving up the stack, from five warehouses to hundreds of agents, copilots and AI apps, each with its own partial view, embedded definitions and private context window. The argument shifts to whose semantics are right and how to keep autonomous systems aligned with one version of reality. Already happening, per two Atlan customers: - "We have 1,000+ Databricks Genie rooms and no way to govern them all. It's like BI sprawl all over again." - "We have all kinds of agentic tools (Sierra, Writer, Google Agentspace, Snowflake Cortex) and none of them talk to each other. I want a common layer of context so I don't need to context-engineer every single one of them." ## Why vertical agents can't solve this ### Execution paths are local, context is global Sitting in the execution path captures decision traces within one workflow, but most decisions draw context from everywhere. A renewal agent proposing a 20% discount pulls from PagerDuty (incident history), Zendesk (escalation threads), Slack (last quarter's VP approval), Salesforce (deal record), Snowflake (usage data) and the semantic layer (definition of "healthy customer"). Every enterprise has a different combination: Salesforce + Zendesk + Snowflake; HubSpot + Intercom + Databricks; a homegrown CRM + ServiceNow + BigQuery. Dozens of agents from dozens of vendors each build their own context silo. To truly capture the context graph, a vertical agent would need 50 to 100+ integrations just for common cases, duplicated across every sales, support, finance and HR agent company. ### The two halves of context Context itself is heterogeneous. Following [Tomasz Tunguz](https://tomtunguz.com/operational-analytical-context-databases/): - **Operational context databases:** SOPs and institutional knowledge (password resets, NDA reviews, options-vesting questions); trade secrets and IP. - **Analytical context databases:** the evolution of semantic layers; definitions and calculations for metrics like revenue or CAC. Semantic layers told AI what data meant; analytical context teaches AI how to reason about it. A renewal decision uses both: the discount exception policy (operational) and how customer health and "at-risk" are calculated (analytical, defined in a semantic layer over a warehouse fed by CRM, support, product analytics and billing). The vertical agent sees the workflow, not the analytical context; the warehouse sees the metrics, not the operational decisions. The context graph must bridge both. ## From execution paths to compounding systems With hundreds of agents, the hard problem is not initial capture but coordination and improvement: how context gets better, stays consistent, and how one agent's learning benefits another. Two foundations: feedback loops and context platforms. ### Context compounds through feedback loops Tunguz: "The key to both operational & analytical context databases isn't the databases themselves. It's the feedback loops within them." The winner gets better at capturing and delivering context over time. Flywheel: accuracy creates trust, trust creates adoption, adoption creates feedback, feedback creates accuracy. A vertical agent can run this only within its workflow. It can't improve shared building blocks: definitions of key terms, entity resolution across systems, metric semantics, cross-domain precedents. A universal context layer runs the flywheel once at the platform level, so every interaction across sales, support, finance and operations improves shared context (the definition of "customer health", Salesforce-contact-to-Zendesk-user resolution, which exceptions set precedents). Compounding lives at the platform layer, not the application layer. ### From context engineering to context platforms Today enterprise AI relies on manual context engineering: forward-deployed AI engineers and agent PMs gathering context and hand-updating system prompts and evals, with every vertical vendor repeating the work per customer. The shift is to [productized context platforms](https://theoryvc.com/blog-posts/from-context-engineering-to-context-platforms), and a key property is that they are customer-owned. Tunguz: enterprises "handed over both data & compute, then watched as the most strategic asset in their business, how they operate, became someone else's leverage." That is why Iceberg exists and open table formats are winning. Institutional decision-making knowledge (tribal knowledge, exception logic, "we always do X because of Y") is even more valuable, and that is what context graphs capture. Enterprises won't hand slices of their operational DNA to a dozen vertical startups; their strategic asset is context, not agents. They will want open, federated context platforms that any agent can read from, humans can govern, and the organization can improve. ## Who will capture the trillion-dollar opportunity? The winner stitches context across workflows, systems and the heterogeneous enterprise stack. Requirements: 1. **Cross-system connectivity:** integrations with hundreds of sources, from warehouses to CRMs to BI tools to communication platforms. 2. **Operational context synthesis:** extracting SOPs and institutional knowledge from logs, tickets, chats and human behavior. 3. **Analytical context management:** governing metric definitions, business entities and semantic relationships. 4. **Context delivery at inference time:** serving the right context to any agent at the moment of decision. 5. **Feedback loops at scale:** improving context continuously across every interaction. 6. **Governance and trust:** ensuring all agents operate on a shared version of reality. This is a platform problem, not an application problem. Companies that already built #1 and #3, connecting Snowflake, Databricks, BigQuery, Salesforce, dbt and Looker, have a structural advantage: they solved heterogeneity and hold the relationship graph tying systems together. The real opportunity is a **universal context layer** that helps all enterprise data and AI systems work together. In a world of heterogeneity the integrator always wins, and after the Iceberg lesson, the platform that lets customers own their context beats platforms that try to own it for them. ## The Great Data Debate The page invites readers to [The Great Data Debate 2026](https://atlan.com/great-data-debate-2026/), where Jaya Gupta (Foundation Capital), Karthik Ravindran (Microsoft), Bob Muglia (Snowflake) and Tony Gentilcore (Glean) debate who is best positioned to win this opportunity. ## About the author Prukalpa Sankar has spent over a decade building at the intersection of data and AI. She started Context and Chaos (formerly Metadata Weekly) four years ago as a space for experts and practitioners to share what works. Forbes 30 Under 30, Fortune 40 Under 40, TED Speaker, and Co-founder & Co-CEO of Atlan. [LinkedIn](https://www.linkedin.com/in/prukalpa/). ## Related reads - [Semantic Layers Failed. Context Graphs Are Next.](https://atlan.com/context-and-chaos/issue/semantic-layers-failed-context-graphs-are-next/) (February 2026) - [Ontologies, Context Graphs, and Semantic Layers: What AI Actually Needs in 2026](https://atlan.com/context-and-chaos/issue/ontologies-context-graphs-and-semantic-layers-what-ai-needs-in-2026/) (January 2026) - [Conceptual Modeling Is the Context Engineering Nobody Is Doing](https://atlan.com/context-and-chaos/issue/conceptual-modeling-is-the-context-engineering-nobody-is-doing/) (April 2026) - [Context Graphs as AI Evaluation Infrastructure](https://atlan.com/context-and-chaos/issue/context-graphs-as-ai-evaluation-infrastructure/) (April 2026) Originally published in the [Context & Chaos newsletter on Substack](https://metadataweekly.substack.com/p/context-graphs-are-a-trillion-dollar). Browse all articles at https://atlan.com/context-and-chaos/.