Context Graphs and the Open System of Record for AI Agents

Foundation Capital recently framed the opportunity clearly: agents do not just need rules. They need access to decision traces that show how rules were applied in reality, where exceptions were granted, and which precedents actually govern outcomes. As those traces accumulate, they form a context graph: a living record of decisions stitched across entities and time.

This page is about two ideas:

1. Context graphs are how agent autonomy compounds inside organizations.
2. The internet also needs an open system of record at the interaction boundary, so decisions across vendors, CDNs, payment rails, and agent frameworks remain auditable and portable.

Originary is building and stewarding that open layer.

Traditional systems of record store objects: tickets, invoices, accounts, documents.

In the agent era, the unit of truth is often a decision:

• which data was used
• which policy version applied
• which exception path was invoked
• who approved a deviation
• what changed as a result

Foundation's key distinction is worth adopting verbatim in spirit:

Agents can follow rules. They get stuck when the real world deviates from rules. Humans resolve those deviations with judgment and organizational memory. Most organizations do not store that memory as a durable artifact.

Foundation argues that enterprises are missing the layer that actually runs them: exceptions, overrides, escalations, and the cross-system context that lives in conversations and ad hoc approvals.

That gap shows up everywhere agents touch production:

This is why “governance alone” does not solve autonomy. The missing ingredient is a queryable record of how decisions were actually made.

A context graph is not “chain-of-thought.” It is not an internal model artifact.

It is a living, queryable record of decision traces, stitched across entities and time, so precedent becomes searchable and automation becomes safer.

The compounding effect is the important part:

• More decisions generate more traces.
• More traces create better precedent.
• Better precedent lets agents handle more cases safely.
• That creates more traces.

This is how autonomy scales without turning into chaos.

This boundary layer is where disputes form and where interoperability matters. It cannot be owned by a single vendor if it is to become durable infrastructure.

A few public shifts show the pressure toward policy, payment, and proof on the open web:

These are important because they establish a shared direction: policy and payment at the protocol edge.

What is still missing is an open, portable proof object that makes those interactions auditable across implementations.

An open system of record should not try to replace payment rails, CDNs, or internal agent platforms.

It should define the evidence object that can reference them.

• Payment standards like x402 define how value is exchanged over HTTP 402.
• Edge systems like Pay Per Crawl define enforcement surfaces and pricing headers.
• Provenance and preference standards can define how origin and consent signals are expressed (where applicable).

The missing layer is a neutral format for “what happened” that can be independently verified later.

To become infrastructure, the boundary-layer record must be:

PEAC Protocol (Programmable Environment for Agent Coordination) targets this boundary layer: policy plus cryptographic receipts for agent interactions.

The core idea is simple:

• Publish machine-readable policy (the “rules” layer).
• Emit cryptographic receipts (the “decision trace event” layer).
• Make verification portable.

This directly complements the context graph thesis:

Below is a realistic, prioritized set of improvements that map to the thesis and to market pressures.

{
  "policy_ref": "hash or version pointer",
  "policy_match": "which rule matched",
  "exception_ref": "opaque reference, not PII",
  "approval_ref": "optional",
  "decision_id": "stable identifier for dedup and linkage"
}

How to position without bluffing:

The constraint is structural: you must sit in, or adjacent to, the execution path.

Foundation explicitly notes that systems-of-agents startups have advantage because they sit in the execution path and can persist decision traces at decision time.

So the GTM should be execution-path-first:

Without centralizing the protocol:

This preserves the open protocol while giving Originary a credible commercial wedge.

Start building with receipts today.