context graph
Captures decision traces and cross-system context to create a queryable record of why enterprise choices were made for AI agents
Last verified:
What is context graph?
Context Graph is a conceptual framework and investment thesis from Foundation Capital describing a new category of enterprise AI infrastructure. It captures decision traces—the exceptions, overrides, precedents, and cross-system context that currently live in Slack threads, deal desk conversations, escalation calls, and people's heads—to create a queryable record of how decisions were made. Unlike traditional systems of record that store only current state (what happened), context graphs store the reasoning behind decisions (why it was allowed to happen).
Key features include capturing decision traces at execution time when agents run, stitching traces across entities and time to make precedent searchable, turning exceptions into encoded knowledge, supporting human-in-the-loop workflows where agents propose and gather context before routing approvals, and building a living record connected by decision events and
context graph pricing
Pricing model: Freemium
This is an investment thesis and conceptual framework from Foundation Capital, not a commercial product. There is no free tier, paid plans, or pricing information. The article describes the context graph category and identifies startups building in this space: Regie (AI-native sales engagement), Maximor (finance automation), PlayerZero (production engineering support), and Arize (observability for agents).
context graph pros
- Captures the 'why' behind decisions, not just the 'what'
- Makes precedent searchable across entities and time
- Turns exceptions into encoded organizational knowledge
- Sits in the execution path to see full context at decision time
- Creates a queryable record of how decisions were made
- Enables auditing and debugging of AI autonomy
- Compounds over time as every automated decision adds traces
- Supports human-in-the-loop workflows from day one
- Captures cross-system synthesis that happens in people's heads
- Records approval chains that happen outside systems
- Preserves tribal knowledge before people leave the company
- Provides structural advantage for systems of agents startups
- Becomes the real source of truth for autonomy
- Allows replaying decision-time state for audit purposes
- Reduces reliance on oral tradition for organizational learning
context graph cons
- Conceptual framework, not a standalone product you can buy
- Requires being in the execution path at commit time
- Incumbents like Salesforce cannot replay decision-time state
- Data warehouses receive data via ETL after decisions are made
- No existing system currently captures decision reasoning as data
- Incumbents may lock down APIs and add egress fees
- Requires instrumenting agent orchestration layer to emit traces
- Full autonomy not required but still complex to implement
Frequently asked questions about context graph
What is a context graph?
A context graph is the accumulated structure formed by decision traces—living records stitched across entities and time so precedent becomes searchable. It is not the model's chain-of-thought, but a queryable record of how decisions were made, explaining not just what happened but why it was allowed to happen.
What are decision traces?
Decision traces capture what happened in a specific case: which definition was used, under which policy version, with what exception, based on which precedent, and what was changed. They include what inputs were gathered across systems, what policy was evaluated, what exception route was invoked, who approved, and what state was written.
How is a context graph different from a system of record?
Traditional systems of record like Salesforce store current state (what the opportunity looks like now). Context graphs store decisions (why it was allowed to happen). Systems of record own canonical data; context graphs own the reasoning connecting data to action that was never treated as data before.
Why can't incumbents like Salesforce build context graphs?
Operational incumbents are siloed and prioritize current state. Salesforce knows what an opportunity looks like now, not what it looked like when the decision was made. When a discount gets approved, the context justifying it isn't preserved. You can't replay decision-time state, so you can't audit, learn from, or use it as precedent.
Why can't data warehouses like Snowflake build context graphs?
Warehouses are in the read path, not the write path. They receive data via ETL after decisions are made. By the time data lands in Snowflake, the decision context is gone. A system that only sees reads after the fact can't be the system of record for decision lineage—it can tell you what happened but not why.
What structural advantage do systems of agents startups have?
They sit in the execution path and see the full context at decision time. When an agent triages an escalation or decides on a discount, the orchestration layer sees what inputs were gathered, what policies applied, what exceptions were granted, and why. Because it executes the workflow, it can capture context at decision time as a first-class record.
What are the three paths for startups building context graphs?
First, replace existing systems of record from day one (like Regie rebuilding sales engagement around agentic execution). Second, replace modules rather than entire systems (like Maximor automating cash/close management while syncing to the ERP). Third, create entirely new systems of record by persisting decision-making traces (like PlayerZero building a context graph for production engineering).
What key signals should founders look for when building?
Two signals apply to all opportunities: high headcount (50 people doing a workflow manually) and exception-heavy decisions (complex logic where precedent matters, like deal desks, underwriting, compliance reviews). One signal points to new system of record opportunities: organizations at the intersection of systems like RevOps, DevOps, Security Ops—'glue functions' where humans carry context software doesn't capture.
What does human-in-the-loop look like in practice?
The agent proposes, gathers context, routes approvals, and records the trace. For example, a renewal agent proposes a 20% discount, pulls incidents from PagerDuty and escalations from Zendesk, routes the exception to Finance, Finance approves, and the workflow captures inputs, approval, and rationale as durable precedent instead of letting it die in Slack.
What startups are mentioned as building context graphs?
Regie.ai builds an AI-native sales engagement platform replacing Outreach/Salesloft. Maximor automates cash, close management, and accounting workflows in finance. PlayerZero automates L2/L3 support while building a context graph for production engineering. Arize builds observability infrastructure for monitoring agent decision quality at scale.