This week I had the opportunity to read the Visa B2AI Report, conducted by Morning Consult with two thousand consumers and 512 executives in the United States. The results didn't surprise me. But the data accurately confirmed something that, in practice, separates what technology can already do from what people actually want it to do for them. And how comfortable they feel doing it. This research reveals exactly that point.
The data reveals a profound asymmetry between companies and consumers and offers a strategic insight that goes beyond enthusiasm for technology. Among the American executives surveyed, 53% would allow AI agents to negotiate prices and terms directly with other agents.
Of that same group, 88% are willing to share pricing and inventory data with corporate AI systems, 77% already use or are piloting AI in their operations, and 71% would adapt products and experiences specifically for AI agents, not humans, for machines.
Imagine a supplier catalog that's no longer a PDF sent by email, but a structured API with dynamic pricing, real-time availability, and pre-programmed negotiation rules, designed to be read by another agent, not a human buyer. That's the world these executives are already building. It's not technological adoption. It's a change in commercial architecture.
Among consumers, 58% are comfortable with AI comparing prices and 55% accept it applying discounts automatically. But only 38% would allow an agent to complete a transaction autonomously, and 60% insist on approving each expense individually.
There's a clear line here: consumers accept AI as a discovery assistant, but resist AI as a decision-making agent. Compare, yes. Buy, not yet.
A necessary caveat regarding the Brazilian context.
The research data reflects the American market. And this distinction matters. It's not just about differences in access to technology or digital maturity. The Brazilian B2B market has its own characteristics that make directly transposing the numbers risky.
Business-to-business trade in Brazil is characterized by significant fragmentation: thousands of small and medium-sized enterprises operate with processes that are still not very automated, a high volume of informal negotiation, and a strong dependence on personal relationships.
At the same time, highly coordinated supply chains coexist with this universe, especially in sectors such as retail, food, agribusiness, and pharmaceuticals, where large players already operate with sophisticated data and systems integration.
There is also the cultural component. In Brazil, commercial negotiation is personal by nature. Delegating this function to an autonomous agent requires a change in mindset that goes beyond the available infrastructure. This doesn't mean that agentic commerce won't arrive in Brazil—it will, and probably with the speed characteristic of our digital leaps, as happened with Pix. But the adoption path will be different, and so will the timing.
Trust is not generic. It has a specific address.
Perhaps the most strategic finding of the research is the least discussed: among American consumers, 36% trust AI systems backed by banks, 35% trust AI enabled by payment networks, compared to only 28% for independent agents.
Eight percentage points may seem like a small number. But in the adoption of financial technologies, they are crucial.
They show that trust is not a generic attribute of AI. It is transferred from institutions that the consumer already knows and respects. Banks and payment networks enter agentic commerce with an asset that no independent agent can buy: a track record built over decades.
For Visa — which has been repositioning itself as an orchestrator of the payment experience, and not just a card network — this research is not neutral. It is also a strategic argument. The company is saying, with data, that its presence in agentic commerce is a differentiator of trust, not just infrastructure.
In Brazil, this logic translates similarly: institutions with a history of relationships and credibility, such as banks, established fintechs, and well-known payment networks, tend to have an advantage in adopting agency solutions over new entrants without this background.
What does this change — and what does it confirm?
Agentic commerce is going to happen. Companies are prepared. The technology exists. Protocols are being standardized in real time. What will determine the pace is the speed at which consumers expand the scope of autonomy they grant to agents, and this depends on transparency, control, and the feeling that the final decision still rests with them.
People are open to AI acting for them, not in their place. It's a subtle distinction—and that's all there is to it.
Acting on behalf of someone implies a clear mandate, defined limits, and the possibility of intervention. Acting in someone's place implies substitution. The consumer accepts the first. They reject the second.
For those working in the payments sector, the implication is straightforward: the next battleground isn't technical. It's about building a governance model that makes the agent trustworthy, not just capable. Robust authentication, a clear audit trail, a simple cancellation mechanism, and above all, the ability for the consumer to know, at any time, what the agent did on their behalf.
The technical infrastructure for agentic commerce is being built in real time. The trust infrastructure is still in its early stages.
And the question remains: when your agent makes a purchase for you, will you want to know how they decided, or will you trust that they knew what they were doing?
* Edson Santos is the founder and partner of Colink Business Consulting, a consultant, advisor, and angel investor with over 25 years of experience in payment methods and financial services. He is the author of "From Barter to Financial Inclusion" and co-author of "Payments 4.0 — The forces that are transforming the Brazilian market."
