AI's trillion-dollar opportunity: Context graphs - Foundation Capital

• The real gap isn't missing data—it's missing decision traces: the "why" behind exceptions, approvals, and cross-system synthesis that humans resolve with judgment but never store as durable artifacts
• "Context graphs" are the structural advantage: startups in the agent orchestration path can capture decision lineage at commit time (what inputs, which policy, whose approval, what precedent) while incumbents only see current state or post-facto ETL
• Incumbents can't build this: Salesforce/Workday are siloed in current state, Snowflake/Databricks are in the read path after decisions happen—neither sits where decisions are made
• Three paths for startups: replace entire systems (Regie rebuilding sales engagement), replace modules (Maximor owning finance workflows), or create new systems of record (PlayerZero capturing production engineering context)
• Build where "glue" functions exist: RevOps, DevOps, SecOps roles signal cross-system workflows where no incumbent owns the truth and decision traces become the moat

The piece introduces "context graphs" as a fundamental reframing of the AI-kills-SaaS debate. While existing systems of record (Salesforce, Workday, SAP) will survive, there's a structural opportunity for an entirely new category: systems of record for decisions, not just objects. The key distinction is between rules (what should happen generally) and decision traces (what happened in this specific case, with full context: which inputs, what policy version, whose approval, what precedent). Most enterprises have never systematically captured decision traces—they live in Slack threads, escalation calls, and tribal knowledge.

The authors argue that "systems of agents" startups have a structural advantage because they sit in the execution path where decisions happen. When an agent orchestrates a workflow (contract review, quote-to-cash, support escalation), it sees the full picture at decision time: what data was pulled from which systems, what policy was evaluated, what exception route was invoked, who approved, and what state was written. If you persist those traces as first-class records, you get a queryable history of how context turned into action—precedent becomes searchable, exceptions compound into organizational memory, and the "why" becomes durable data. Incumbents can't build this: operational systems like Salesforce only store current state (not decision-time context), while warehouse players like Snowflake/Databricks are in the read path after decisions are made via ETL.

The piece outlines three paths for startups: replace entire systems from day one (Regie rebuilding sales engagement for AI-native teams), replace specific modules while syncing to incumbents (Maximor owning finance workflows while the ERP remains the ledger), or create entirely new systems of record (PlayerZero capturing production engineering decision traces that no existing system stores). The key signals for where to build: high headcount workflows, exception-heavy decisions, and especially "glue" functions (RevOps, DevOps, SecOps) that exist precisely because no single system owns the cross-functional workflow. These roles carry context that software doesn't capture—automating them means capturing a category of truth that only becomes visible once agents sit in the workflow.