Loyal Agents

The Initiative

The Next Era of the Internet Will Be Run by Agents

AI agents are moving past answering questions into taking action on our behalf. Whoever an agent is loyal to determines who wins from that shift.

New to the term? Read our definition of a loyal agent.

The shift underway

From Answers to Authority

From Chat to Action

Today’s agents book, buy, cancel, and negotiate, often with real money and consequences attached.

From One-Off to Ongoing

Agents carry preferences, history, and constraints across sessions. That persistent memory needs clear rules about control.

From Platform-Bound to Cross-Platform

Shared standards for identity, authorization, and accountability are essential when agents act across services.

What’s at stake

A System-Level Question

Economic

Standards will shape market structure, competitive dynamics, and who captures trillions in value.

Civic

Platform-loyal agents can entrench power asymmetries and quietly erode individual autonomy.

Trust

Consumers and businesses need transparent, auditable agents whose loyalty can be verified.

The trust gap today

Hidden Incentives Are Not an Edge Case

“A travel agent favors the itinerary paying the highest affiliate commission.”
“A subscription agent ‘forgets’ to cancel because the platform benefits from inaction.”
“A shopping agent cannot explain why it recommended one product over another.”

Comparison of a shopping agent with a hidden incentive and a loyal agent answering the same request.

Same request, two agents.

"Find me noise-canceling headphones under $150. I mostly use them for calls."

Agent with a hidden incentive

Recommends

Headphones A · $249

Why: "Our top pick." It pays the highest commission.

Disclosure: None.

Your limits: $99 over your budget.

Loyal agent

Recommends

Headphones B · $129

Why: Fits your budget and rates best for call clarity.

Disclosure: Earns a referral fee here, and shows it didn't change the ranking.

Your limits: Asks before spending more than $150.

Why now

The Rules Can Still Be Written

Agentic AI is capable enough to matter, but the standards governing its behavior toward users do not yet exist. This is the moment to act before harms become systemic.