Provenance receipts make each autonomous decision traceable back to evidence a human operator would accept.
Where did the claim come from: CRM record, call transcript, support ticket, product usage event, human approval, or external signal?
// TRUST DENSITY · AGENT OPS · June 19, 2026
Trust Density Needs Provenance Receipts Before Agents Copy Each Other
When agents reuse each other’s conclusions, trust stops being a feeling and becomes a receipt chain.
Machine teams can multiply work faster than managers can remember where the original truth came from.
Operators can let agents collaborate without letting recycled assumptions become revenue policy.
Agent teams do not create trust by agreeing with each other.
In a human team, repeated agreement can be useful because people bring different context, memory, and accountability. In an agent team, repeated agreement can be a hall of mirrors.
One worker enriches an account from a stale CRM field. Another summarizes that enrichment into a segment note. A third writes an email because the segment note says the account is expanding. By the time the message reaches a customer, three systems appear to support the action.
But there may have been only one source, and it may have been wrong before the first agent touched it.
A provenance receipt is the minimum unit of operational trust.
Who owns the source and who can correct it when the agent discovers conflict or missing context?
When was the evidence last verified, and how long is it safe for this workflow?
What is the agent allowed to do with this evidence: draft, route, write to system, contact customer, or escalate?
The risky moment is not hallucination. It is second-hand certainty.
Most operators already watch for fake facts in single-agent output. The harder problem is copied certainty. A claim that was acceptable for one low-risk workflow becomes dangerous when another agent imports it into a higher-risk workflow.
Example: a pricing note can be safe for internal account research and unsafe for a customer-facing renewal email. A churn-risk label can be safe for queue prioritization and unsafe for forecast changes. A competitor mention can be safe for analyst notes and unsafe for automated outbound.
Without provenance, the second agent sees a clean conclusion. The operator loses the messy evidence boundary that made the first conclusion tolerable.
The receipt contract
Claim
The exact fact, judgment, or recommendation the agent is passing forward. No vague summaries like “high intent” without the underlying signals.
Evidence
Links or identifiers for the records used, including timestamps, system names, and freshness windows.
Allowed use
The permitted next actions: read-only analysis, draft-only work, internal routing, system write, customer contact, or financial decision.
Owner
The human or system owner accountable for correcting the evidence and handling exceptions.
Expiry
The point where another agent must refresh, quarantine, or escalate instead of reusing the claim.
The metric is unsupported reuse rate.
Track every time an agent uses another agent’s output without carrying the source receipt forward.
This metric matters because machine labor scales through reuse. If reuse is unsupported, every workflow improvement increases the surface area of untraceable decisions. The org may feel faster while its truth system gets weaker.
Healthy trust density means more autonomous work can happen per verified fact, but only while the fact remains visible, scoped, and fresh.
If an agent cannot show the receipt, it is not using trust. It is spending it.
Operator checks
Three signs your agents need provenance receipts
Teams discuss agent-written briefs more often than the underlying CRM, ticket, call, or usage evidence.
A low-risk analytical label quietly becomes the basis for routing, forecasting, customer outreach, or pricing action.
When a fact is fixed in one system, downstream agent plans and drafts keep using the old version.
Install receipts before adding another collaborator agent.
Pick a workflow where one agent already consumes another agent’s research, scoring, routing, or draft output.
Require source, owner, timestamp, scope, and expiry with every claim that crosses the handoff.
If the receipt is missing or expired, the receiving agent must refresh, quarantine, or ask instead of continuing.
Audit every reversed or corrected action and ask which receipt failed to travel with the work.
Concrete takeaway for this week
Before letting agents copy each other’s conclusions, define the provenance receipt that must travel with every reusable claim.
- Name the source: the record or signal behind the claim.
- Name the owner: who can correct the source and resolve conflicts.
- Name the timestamp: when the evidence was last safe to use.
- Name the scope: what actions this evidence permits.
- Name the expiry: when reuse must stop.
The blunt rule: autonomous companies do not get safer because more agents agree. They get safer when every agreement can prove where its truth came from.