AI is still being judged like software someone talks to. Which model gives the best answer? Which assistant writes the cleanest summary? Which interface feels most natural?
That can’t last.
The next phase of AI is not just about generating answers. It is about taking action. Agents will book travel, compare subscription bundles, apply discounts, unlock gated pricing, validate professional credentials, confirm eligibility for benefits, verify age or residency when relevant, route consumers to the right experience, and complete transactions on behalf of consumers.
That shift changes the trust problem. When AI makes a recommendation, a best guess can be useful. When AI executes, a best guess isn’t good enough.
Consumers are already drawing that line. Nearly half of business buyers say they would work with an AI agent if it meant faster service, but only 17% of customers are comfortable with an agent making financial decisions for them. That gap captures the shift now underway: people may welcome AI when it removes friction, but they expect a higher standard when it starts making decisions that affect money, access, or eligibility.
A model can infer that someone may be a student, a university graduate, a licensed professional, an association member, an educator, age-qualified for a product, service, or offer, or a resident of a specific state, country, or region. It can scan public information and make a probability-based assessment.
But it shouldn’t redeem an offer, grant access, or confirm eligibility, validate an age-based experience, or apply a residency-based rule based on something scraped from the open web.
Execution requires confirmed facts tied to sources of record.
The new action layer needs a truth layer
Commerce is full of decisions that depend on real-world attributes. Is this person currently enrolled? Do they belong to the organization they claim to belong to? Are they licensed in a specific field?
Are they eligible for this benefit? Are they old enough to access this product or service? Do they live in a geography where a certain offer, benefit, or experience applies? Did they consent to use this credential in this transaction?
Is the claim still current?
Those questions become much more important when software acts at scale. A consumer may apply for one offer at one checkout. An agent can search, compare, and attempt to redeem across many brands in seconds.
That creates enormous potential for convenience, but it also creates risk. If brands make every action too hard to complete, legitimate customers face unnecessary friction. If they make high-value offers too easy to claim, they invite abuse and weaken the program’s value.
The answer isn’t more friction. It’s better verification. AI agents need a way to determine which claims are true, which are current, and which the consumer has consented to share.
“Probably eligible” isn’t a strong enough foundation for automated commerce.
Special offers and discount programs are one clear example, but they are not the outer boundary. The same infrastructure can support many other AI-driven decisions where a verified attribute changes what action should happen next. Age can determine whether a consumer is eligible to access a product, service, or experience. Residency can determine which benefit, offer, content, or service path applies.
Professional credentials can determine whether someone should receive access to a specialized resource. Each use case is different, but the underlying requirement is the same: AI needs confirmed, permissioned facts before it acts.
The model is not the moat
AI competition is often framed around the model itself. That framing made sense when the primary question was which system could generate the best output. But as models improve and web-scale data becomes more widely available, durable differentiation will depend on the quality of the data behind the action.
That’s especially true in high-trust use cases. Some will involve offers and benefits. Others will involve access, personalization, product availability, customer routing, or eligibility for a specific experience.
The open web can provide context, language, product details, reviews, and broad signals. It can’t reliably confirm whether a person has a specific real-world attribute at a specific point in time.
Eligibility isn’t a sentiment. Licensure isn’t a prediction. Membership isn’t a behavioral segment. Enrollment isn’t a lookalike audience.
Age qualification is not a guess. Residency is not something an agent should assume from browsing behavior. These are claims that need to be verified against authoritative sources, kept current, and shared in permissioned, controlled ways.
For brands, this is where AI becomes both more powerful and more demanding. The better the agent, the more consequential its actions become. The more consequential the action, the less room there is for guesswork.
This is why authoritative data becomes more valuable as AI becomes more capable. Companies that can confirm real-world attributes and eligibility with accuracy will have an advantage that can’t be recreated by simply adding another model or interface.
Permission is part of the infrastructure
There is another requirement that should not be treated as an afterthought: the consumer needs to remain in control.
The infrastructure isn’t keeping pace with the ambition. Nine in ten organizations say AI has expanded the scope of their privacy programs, yet only a small minority describe their AI governance efforts as mature and proactive. That gap matters because agentic commerce will depend not only on what AI can do, but on whether the data and permissions behind each action can be trusted.
The old model of proving eligibility often asked people to hand over more information than the transaction required. That approach can create friction for the customer and risk for the enterprise.
The better model is more precise. A person should be able to show that a claim is true without giving away unnecessary information. For example, they should be able to prove they are old enough for a product or service without exposing every detail on an identity document. They should be able to confirm residency for a relevant benefit without turning that interaction into a broad data collection exercise.
They should be able to use a credential for a specific purpose, in a specific context, with a clear understanding of what’s being shared and why.
In agentic commerce, that permissioning layer becomes essential. An AI agent shouldn’t simply possess data and use it wherever it can. It should understand which verified claims are available, what actions they support, and what permissions govern their use.
That’s the difference between automation that feels helpful and automation that feels invasive.
Verifiable truth will decide what AI can do
The next stage of AI won’t be defined only by better answers. Better actions will define it.
For brands, that means preparing for a world where AI agents aren’t just influencing the customer journey but participating in it. Offers, benefits, memberships, credentials, age-based rules, residency-based experiences, and access rules need to be machine-readable without losing the consent, context, and control that consumers expect.
For AI systems, it means recognizing that some decisions can’t be made through inference. Some actions require a verified yes-or-no. Others require a verified attribute that tells the system which action is appropriate.
For consumers, it means AI can become more useful without asking them to surrender more personal data than necessary.
The foundational layer of agentic commerce won’t be the model alone. It will be the integrity of the data that determines which actions are allowed. As more authoritative consumer attributes become available to AI systems in controlled, permissioned ways, the value of verification will expand well beyond traditional offer programs.
It will become part of how AI decides what it can do, what it should not do, and what it needs to confirm before acting.
As AI moves from interface to execution, verifiable truth becomes infrastructure.
