The Context Graph Opportunity Everyone’s Missing

Foundation Capital and Dharmesh Shah are both right. They're also both missing something.

Foundation Capital dropped a piece last month calling context graphs “AI’s trillion-dollar opportunity.” The thesis is compelling: our systems capture what happened, but not why. The reasoning behind decisions—what inputs were considered, what exceptions were granted, who approved what—lives scattered across Slack threads, meeting transcripts, and people’s heads. It resets every time someone leaves.

Context graphs are supposed to fix this. Capture decision traces as agents work. Build a queryable history of real-world precedents. Stop relying on tribal knowledge and start learning from accumulated actions.

The theory is elegant. Dharmesh Shah’s response was more grounded.

Writing at simple.ai, the HubSpot co-founder called context graphs “a beautiful idea that will matter eventually”—then pointed out that most companies are nowhere near ready. They’re still struggling with basic data unification. They’re early in AI adoption, figuring out if an assistant can handle tier-1 support. Agents, whose activity is supposed to generate the decision traces that populate context graphs, are themselves barely deployed at scale.

His analogy stuck with me: asking companies to capture decision traces right now is like asking someone to install a three-car garage when they don’t own a single car.

He’s right. And he’s also missing something.

---

The implicit assumption in this entire debate is that companies build context graphs for themselves. Foundation Capital frames the opportunity as startups building “systems of agents” that capture context at decision time. Dharmesh frames the readiness problem as companies needing to deploy agents before they can instrument them.

But what if the context graph isn’t something companies build? What if it’s something built for them—by whoever already sits in the decision-making seat?

I’ve spent 17 years building a strategic finance firm embedded inside venture and growth-stage companies. Not fractional. Not outsourced. Embedded—multi-year partnerships where our teams serve as the finance function, handling everything from month-end close to board reporting to the judgment calls in between.

We’ve worked with over 1,400 companies. Twenty-two became unicorns—with at least another half-dozen in the current portfolio who’ve crossed that threshold in revenue but haven’t raised at the valuation to make it official. The accumulated pattern recognition across that portfolio is substantial. But here’s what’s become clear watching this context graph conversation unfold: the most valuable thing we’ve built isn’t a database or a platform. It’s the structural position to see how financial decisions actually get made—across hundreds of companies simultaneously.

Every month, our teams close books, prepare board decks, and navigate the gray areas where policy meets reality. Should this expense be capitalized? Does this contract trigger rev rec treatment? What’s the right reserve against this receivable? These aren’t questions with algorithmic answers. They’re judgment calls informed by context—context that currently lives in our people’s heads and disperses the moment an engagement ends.

We’re not unique in having this problem. We’re unique in having it at scale.

---

Arvind Jain at Glean offered a refinement to the context graph thesis that I think is underappreciated. You can’t reliably capture the “why,” he argued—that’s a thinking step that usually resides in someone’s head. But you can capture the “how”: recurring steps, approval patterns, collaboration sequences, escalation triggers. Over many cycles, those process traces approximate the why. You infer rationale from patterns in how work repeatedly gets done.

This reframe matters because it changes what you need to build a useful context graph. You don’t need to extract reasoning from Slack threads retroactively. You need to be present at decision time, consistently, across enough similar decisions that patterns emerge.

The Foundation Capital piece makes this point about agents: “Because it’s executing the workflow, it can capture that context at decision time—not after the fact via ETL, but in the moment, as a first-class record.”

But agents aren’t the only things that execute workflows at decision time. Humans do too—specifically, humans whose job is to make financial decisions for companies that have chosen to embed rather than build.

---

Here’s what I think the context graph debate is missing: the adoption problem that concerns Dharmesh might not be solvable at the individual company level. A Series B startup isn’t going to build decision-capture infrastructure—they’re trying to hit milestones and not run out of money. The investment doesn’t pencil at their scale.

But aggregate hundreds of those companies into a single ecosystem processing $4 billion annually, and the economics flip. You build the infrastructure once and deploy it across every engagement. The subscale company gets access to capabilities they could never justify alone. The aggregator gets data density and pattern recognition that no single company—regardless of size—could generate internally.

This is the piece no one in the current debate is talking about. The context graph opportunity might not be captured by the startups building agent orchestration layers. It might not be defended by the incumbents protecting systems of record. It might accrue to whoever already sits in the execution path across enough companies to see patterns that no single company could see alone.

I don’t know exactly what this means yet. We’re early in figuring out how to systematize what we’ve been doing implicitly for years—how to capture decision traces at the moment they happen, how to make accumulated judgment queryable, how to turn portfolio-scale pattern recognition into something more durable than institutional memory.

But I’m increasingly convinced the “who builds the context graph” question has an answer no one is considering: not the companies themselves, and not the software vendors serving them, but the embedded partners who’ve been making decisions alongside them all along.

The three-car garage problem is real, and subscale companies don’t have cars. But some of us have been building the pit crew for 1,400 teams over 17 years—and we’re just now realizing what all those laps taught us.

---

I’m Chris Fenster. Founder, builder, occasional overthinker. 17+ years helping venture and growth-stage companies grow up. I write here to figure out what I think.

---