№ 07·0507 · AI Agent System3 min read · Section 5 of 5
7.5 How the Agent enters the business loop
Task binding, human adoption, the evidence chain, and PoB — the path that turns Agent output into a settleable result, with worked examples.
Updated
7.5 · Into the business loop
Adoption and proof are the gate of value. Generation is not.
An Agent that runs a demo only proves that it can generate. The WCN loop requires proof that it advanced a result. So the system records which output is bound to which task, who adopted it, what changed after adoption, and whether the final evidence passed Proof. Without that chain, it is one more chat window.
What this page doesDefines the closing conditions, the workflow, and the link to PoB
Core pathOutput, then adoption, then result, then Proof, then PoB
You will learnTask binding, the human adoption event, and traceability
The main path into the loop
A task is dispatched with its type and acceptance criteria. The Agent executes within its authorization. The structured output is stored. The node owner adopts, rejects, or revises it. The state machine advances the deal or task. The result evidence is submitted. The Proof Ledger reviews it. A PoB record is generated. The entry reaches settlement and the record.
The Agent's value is as a traceable driver of results. The system separates "ten pages were generated" from "three were adopted and led to a meeting held, a term changed, or a payment condition met".
What Agent output needs to reach Proof
Bound to a taskEach output is tied to a task handle. Loose results drifting in private chats or offline documents are not eligible.
Input and version recordedThe model name and version, the parameters, the retrieved fragment, and the key attachment hash — enough to reproduce the output and to collect evidence in a dispute.
Human adoption recordedAn explicit event: who adopted it, when, and any revision. Unlike an implicit copy-and-paste, this is structured and kept.
Linked to a resultTraceable to a deal, a milestone, or a business event. Without that link, attribution and settlement cannot be done.
If any condition is missing, the output is treated as tool-generated and is not counted into PoB. The reason matches the one in accounting: an item without a reference and a working paper cannot enter the books.
Three worked examples
Diligence material
Task: the risk section of a Series A package. Project Diligence pulls comparable risk disclosures and drafts the section with sources. The node revises and adopts it, then uploads the revised file hash. Proof links it to the final draft used in the investor meeting, and a PoB record follows once the meeting closes.
Evidence review
An evidence package is submitted for a completed task. The PoB Evidence Validator checks it against the required fields and flags two gaps. The node fills the gaps, and only the completed package goes to the Proof Ledger; the Validator's first pass does not pass it on its own.
Monitoring follow-up
A monitoring note reports an anomaly. Outreach and Meeting Summary drafts an internal task note. The risk owner adopts one and creates a follow-up task. The receipt enters the evidence package, and the entry moves to PoB once Proof passes.
The counter-case: busy, with no loop
An open, unbounded run produces many notes with no task and no owner. That is noise.
An Agent drafts a polished email that is not sent from a WCN task and carries no adoption tag. That cannot be attributed.
An Agent suggests a step, the node forwards it by word of mouth, and the system holds no state. That cannot be verified.
What this step decides
Tool, or execution layerAgent work becomes the network's execution layer only after it enters the loop. Before that, it is no different from any model wrapper.
Settleable, or notWithout a Proof chain there is no PoB, and incentives and shares have no evidence to rest on.
Manageable, or notAdoption rate, reverted adoptions, and model comparisons can be measured, so an Agent configuration can be upgraded or suspended on evidence.
The WCN differenceThe point is not that it connects to a model. It is that machine output, across a multi-node business network, reaches one shared language of evidence and settlement.
A test for acceptance: in a dispute, can the logs explain in half an hour who asked the Agent to do what, what was adopted, and what the final business result was. If they can, the loop is sound.
