Context Graphs: AI's Missing Decision Layer
TL;DR:
-
Context-graph capture why decisions were made, not just what happened
A living record of decision traces stitched across entities over time, capturing not just what happened but why it happened. Context graphs make decision lineage queryable, turning exceptions into precedent and enabling AI agents to understand how organizations actually operate beyond formal policies. - They bridge the gap between formal
Systems-of-record and tribal knowledge in Slack threads
Enterprise software applications that serve as the authoritative source of truth for specific business data. Examples include CRMs for customer data, ERPs for financial data, and HRIS for employee information. These systems capture what happened but often miss why decisions were made. - Foundation Capital calls this a trillion-dollar opportunity as
Ai-agent need
An autonomous AI system that can execute tasks, make decisions, and take actions across multiple systems without constant human intervention. Agents use language models to understand goals and interact with tools, APIs, and data sources to accomplish work.Decision-trace to scale autonomy
A record of why a specific decision was made, including the context, inputs considered, exceptions granted, approvals obtained, and precedents followed. Decision traces capture the reasoning behind actions, not just the final outcome, and form the building blocks of context graphs. - The debate: are systems of record enough (Jamin Ball), or do we need a new data category (Foundation Capital)?
Here’s the problem with enterprise data right now.
Your
Customer Relationship Management - Software that manages a company’s interactions with customers and prospects. CRMs track sales pipelines, customer communications, deal stages, and revenue data, serving as the system of record for customer information.
It has no idea why the discount was approved. That’s the gap.
The why lives in:
- Slack threads where someone mentioned the customer threatened to churn
- A hallway conversation where your boss gave you a thumbs up
- That one email thread where finance agreed to the exception
- Your head, because you’ve seen this pattern before with healthcare companies
This is what Foundation Capital calls “decision traces” · the missing layer that actually runs enterprises.
A context graph is a queryable record of decision traces stitched across entities over time.
Think of it like this: systems of record capture state (the 20% discount exists), but context graphs capture decision lineage (why the discount was allowed).
The distinction matters more as AI agents take on actual work.
Agents don’t just need data. They need to understand precedent, exceptions, approval chains, and cross-system reasoning that led to past decisions.
From the Foundation Capital essay:
“Agents don’t just need rules. They need access to the decision traces that show how rules were applied in the past, where exceptions were granted, how conflicts were resolved, who approved what, and which precedents actually govern reality.”
This is the what vs why gap in practice.
The conversation started with Jamin Ball’s essay “Long Live Systems of Record” in December 2025.
Ball’s point: before agents, you need canonical data. If your sales team and finance team disagree on what ARR means, agents will fail.
He writes:
“Anyone who has spent time inside a large company knows how messy this gets in practice. Take something as simple as what is our ARR? Ask sales and you will get one number. Ask finance and you get another with a different set of exclusions and adjustments.”
Ball argues that fixing this, rationalizing what the correct answer is and where it lives, is unglamorous but critical work.
Foundation Capital agrees with this. But they add something crucial.
Even if you reconcile all your systems of record, you’re still missing an entire category of information.
From their essay:
“Ball’s framing assumes the data agents need already lives somewhere, and agents just need better access to it. That’s half the picture. The other half is the missing layer that actually runs enterprises · the decision traces, exceptions, overrides, precedents, and cross-system context that currently lives in Slack threads, deal conversations, escalation calls, and people’s heads.”
This is where context graphs come in.
Let’s make this real with examples from the Foundation Capital essay.
Exception Logic in People’s Heads
“We always give healthcare companies an extra 10% because their procurement cycles are brutal.”
That’s not in your CRM. It’s tribal knowledge passed down through onboarding conversations.
A context graph would capture this as a decision trace: healthcare vertical + long procurement cycle = discount exception pattern.
Precedent from Past Decisions
“We structured a similar deal for Company X last quarter. We should be consistent.”
This lives in someone’s memory or buried in old email threads.
A context graph makes precedent searchable. Your renewal agent can query: “Show me similar deals in this vertical from the past 6 months.”
Cross-System Synthesis
A human looks at:
- Data in Salesforce (account health score)
- Open escalations in Zendesk (3 critical tickets)
- Slack thread where someone flagged churn risk
- PagerDuty incidents (2 SEV1 outages last month)
They synthesize all of this in their head and decide to escalate to tier 3 support.
The record only says “escalated to tier 3.”
The context graph has the full decision trace · what inputs were considered, how they were weighted, who made the call.
Approval Chains Outside Systems
Your boss walks by your desk. You ask if you can add 5% to the discount. They give you a thumbs up and keep walking.
The CRM only shows the final price. Not who approved the deviation or why.
Context graphs capture these informal approval chains if agents are in the execution path.
Here’s how Foundation Capital illustrates this in practice.
A renewal agent proposes a 20% discount.
Policy caps renewals at 10% unless a service impact exemption is approved.
The agent:
- Pulls three SEV1 incidents from PagerDuty
- Finds an open “cancel unless fixed” escalation in Zendesk
- Locates the prior renewal thread where a VP approved a similar exemption last quarter
- Routes the exception to finance with all this context
- Finance approves
The CRM ends up with one fact: 20% discount.
The context graph has:
- Why the exception was needed (service impact)
- What precedent existed (last quarter’s VP approval)
- Who approved this time (finance)
- What evidence justified it (SEV1 incidents + Zendesk escalation)
Over time, these records form patterns.
The agent learns: service impact + precedent + escalation severity = discount exception path.
Next time, it can route similar cases automatically because the decision trace is queryable data.
One of the most interesting follow-ups came from the Cogent Enterprise substack.
Their argument: don’t predefine context graph schemas.
Traditional
A structured representation of knowledge using entities and their relationships, typically stored as a graph database. Knowledge graphs help AI systems understand connections between concepts, but they require predefined schemas unlike context graphs which emerge from agent execution traces.
Context graphs invert this.
Agents act as “informed walkers” through your decision landscape. As they solve problems · traversing APIs, querying docs, reviewing past tickets · they discover the organizational ontology on the fly.
Each trajectory leaves a trace:
- Which systems were touched together
- Which data points co-occurred in decision chains
- How conflicts were resolved
Accumulate thousands of these walks and the organizational schema reveals itself from actual usage patterns rather than predetermined assumptions.
Example: imagine your company has a policy that in practice gets broken almost every time.
The exception isn’t actually the exception. It’s just the rule in practice.
If you predefined your context graph based on formal policy, you’d miss this. But if agents discover it through actual execution traces, they’ll surface the policy-in-practice rather than the policy-on-paper.
This is powerful.
Box CEO Aaron Levie wrote about this in his essay “The Era of Context.”
His key point: if everyone has access to the same AI talent (these agentic superintelligences), what differentiates companies?
Context.
Levie writes:
“We imagined that AI systems would adapt to how we work. But it turns out due to their extreme power and inherent limitations, we will instead adapt to how they work. This means we will have to optimize our organizations and workflows to best enable context for agents to be successful.”
The individual contributor of today becomes the manager of agents in the future.
Their responsibilities:
- Providing oversight and escalation paths
- Coordinating work between agents
- Guiding agents with the right context
- Making judgment calls that break patterns
Decision traces · the what context graphs capture · are the most uniquely human part of how work gets done.
They’re the decisions that break rules or break out of patterns. The judgment calls that respond to reality as it presents itself, not as you imagined it.
Foundation Capital calls context graphs a trillion-dollar opportunity.
Here’s why:
Early mover advantage in decision trace capture. If agents naturally create context graphs as they execute work, companies that deploy agents early will have richer decision histories.
Context as competitive moat. You can copy someone’s tech stack. You can’t copy their decade of decision traces showing how they actually operate.
New infrastructure category. Just like data warehouses emerged to rationalize operational data, context graph platforms will emerge to capture and query decision lineage.
Governance and observability. As agents take on more autonomy, being able to audit “why did the agent do this?” becomes critical. Context graphs make autonomy debuggable. If you’re concerned about AI agent security risks, decision trace audit logs are part of the answer.
This is context engineering · designing systems to get agents access to the right data and ensuring they can interoperate on that data.
It’s one of the key predictions for enterprise AI in 2026.
Let me bring this back to the core insight.
Systems of record are good at state. This deal closed. This ticket was escalated. This price was set.
They’re bad at decision lineage. Why was the deal structured this way? Why was this ticket escalated? Why was this price approved?
Context graphs make the “why” queryable.
They turn exceptions into precedent. They make approval chains visible. They capture the cross-system reasoning that humans currently do in their heads.
As agents scale, this becomes the layer that actually runs enterprises.
Not just the data. The decision traces.
If you’re building enterprise AI systems in 2026, start thinking about how you’ll capture decision context · not just operational state.
That’s the missing layer.
- AI’s Trillion-Dollar Opportunity: Context Graphs · Foundation Capital (Jay Gupta, Ashu Garg)
- Long Live Systems of Record · Jamin Ball / Clouded Judgement
- Box CEO Aaron Levie on AI’s Era of Context · TechCrunch interview