---
title: The agent credit bureau
subtitle: A model for consumer-directed, comparable reporting about AI product performance.
kind: Research note
date: 2026-06-11
author: Ron Bronson
revised: 2026-10-16
---

A personal history of AI work can help someone decide how much authority to give an assistant. The same histories, pooled, might let people compare consumer AI products on evidence drawn from actual tasks.

## Combining customer records to compare AI products

A consumer may know that a particular assistant repeatedly failed to complete a booking or required extensive corrections to a draft. That evidence is useful locally. Comparing it across users requires consistent definitions of the work, its difficulty, and the outcome.

A proposed agent credit bureau would receive opt-in contributions from consumer-controlled records and build comparable product-level assessments. Its subjects would be recognizable products such as Claude, ChatGPT, and Gemini, across task domains. Subagents would not need persistent identities of their own.

## Satisfaction ratings and edits can misstate performance

A submitted email, an accepted email, and a delivered email are distinct states. An edited document is not necessarily an inaccurate document. Ratings derived from satisfaction alone can misstate performance, while verified outcomes may be sparse or delayed.

A shared report therefore needs the source of each claim, a way to identify comparable tasks, and enough observations to report uncertainty. It must also account for changes to the product over time. A rapid shift in scores could reflect a change in who is using the service or what tasks they attempt.

## Customers retain control over agent permissions

An aggregate rating can inform a person's decision and should never expand an assistant's permissions on their behalf. The consumer's local mandate sets the authorization boundary.

The record should be exportable, and participation in a pooled index should be voluntary. A shared report also needs rules against rewarding a provider for purchasing favorable treatment.

## Who would use comparable AI product reports?

Consumers may use reporting to choose a product for a task. Organizations purchasing agent services may use evidence about reliability by task and outcome. Both possibilities require evaluation before anyone can claim a viable reporting market.

Vloora's research treats this as a possible intelligence business built on transaction records. The underlying brokerage should still be useful to an individual if no reporting network forms.

## Sources

- Vloora, Agent credit reporting research brief, updated October 16, 2026, on commercial possibilities, evidence, and credit artifacts
- Vloora, [The customer's side of the counter](/writing/the-customers-side-of-the-counter/), October 4, 2026, on evaluating agent results from the consumer's position