A familiar story is playing out in payments, we are obsessing over a new user interface while the real change is happening in who, or what, is doing the buying. “AI shopping” sounds like a gimmick until you notice the quiet shift in language from assistants to agents, from suggesting to acting. When an AI can search, compare, negotiate, and complete a purchase, the checkout is no longer a human moment. It becomes a machine-to-machine trust event.
Mastercard’s public push to define standards for “agentic commerce” is a tell. Payments networks do not usually rush into cultural trends unless they smell a structural change in behaviour. They are framing a world where software purchases on your behalf, and they are explicitly anchoring that world in security, transparency, and accountability.
Most people still think payments is the final step of a transaction. You choose, then you pay, then the merchant fulfils. That model worked when commerce was a set of discrete, visible moments. In an agentic world, paying becomes a continuous capability, like location services on your phone. It is called when needed, sometimes without you noticing, because the choice has already been delegated.
That delegation changes everything. In traditional ecommerce, the biggest risks were stolen credentials and fraudulent card use. In agentic commerce, the risk surface expands into authorisation chains:
These are not abstract questions. They are operational requirements for scale.
This is why Mastercard is talking about “agentic tokens” and building a trusted payments layer that sits inside the agent environment, not bolted on afterwards. The subtext is blunt, if AI agents become a meaningful share of shopping activity, the payments industry needs a new grammar for consent. Mastercard even positions its work as standards building across partners, rather than a proprietary land grab, which is a tacit admission that trust only scales when it is interoperable.
Here is a useful way to think about it. Agentic payments is not simply a new checkout. It is a new form of identity. The “payer” is no longer a person typing details, it is a delegated identity acting under constraints. The customer experience becomes less about the click and more about the configuration of autonomy, what you allow your agent to do, how you review it, how you revoke it, how disputes are handled when something goes wrong.
The industry has a habit of learning security through pain. We launch, we grow, we patch. That pattern fails in an agentic world because the first large breach will not look like “someone stole my card details”. It will look like “my agent was manipulated”, and the line between user error and system failure will be disputed in public. That is how trust collapses.
So the narrative must change. Agentic commerce is not a novelty, it is a governance challenge disguised as a convenience feature. It requires standards for four things.
First, permissioning, granular mandates that are easy for humans to set and easy for machines to enforce. Second, proof of intent, a way to demonstrate that the action is consistent with what the human authorised, not merely that a credential was present. Third, liability, clear rules on who carries loss when the agent misbehaves, the merchant, the platform, the payments provider, or the consumer. Fourth, dispute rails, because in a delegated world, refunds and reversals will be part of the social contract, not an edge case.
Payment networks pushing standardisation early is encouraging. But standards are not enough if they remain an industry conversation. Merchants need implementation patterns, regulators need assurance that consumer protection is not optional, and the wider ecosystem needs a shared vocabulary for what “authorised” means in an agentic context.
Agentic commerce will not land evenly across regions. In many African markets, the constraint is not consumer appetite, it is merchant enablement, interoperability, and trust. The opportunity is enormous because agentic journeys can reduce friction where infrastructure is uneven. If an agent can handle complex flows, price comparisons, delivery options, and payment selection, it can make digital commerce feel simpler, particularly for first time online buyers or small merchants with limited digital capability.
But there is a risk. If the trust architecture is imported wholesale from high income markets without adaptation, it may fail under local realities, identity systems vary, addressability varies, dispute resolution norms differ, and fraud patterns differ. This creates a strategic opening for African fintech’s and banks to shape local trust layers that reflect how commerce actually works on the ground.
South Africa is especially well positioned. It has sophisticated financial institutions, strong mobile penetration, and a regulatory ecosystem that is increasingly engaged with modernisation and governance. The country can become a proving ground for agentic commerce that is both innovative and safe, particularly if the ecosystem builds clear consent models, merchant tooling, and rapid dispute handling that protects trust.
Agentic commerce is not about AI taking over shopping. It is about commerce becoming software native. The organisations that treat this as a standards and trust project will define the rails that everyone else builds on. The organisations that treat it as a marketing trend will discover, brutally and publicly, that convenience without accountability does not scale.
Contact Us
VUKA is the trusted media partner to key professionals, policy makers, suppliers and
manufacturers. We provide unparalleled opportunities for industry-wide connection.
