For most of digital commerce, the person making a purchase has remained at the centre of the transaction.
A customer searches for a product, compares the options, approves the payment and authenticates it using a PIN, password or one-time code.
Artificial intelligence could disrupt that sequence.
AI agents are evolving from tools that recommend products into systems capable of acting on a user’s behalf. An agent could monitor prices, reorder household supplies, renew subscriptions, book travel or pay an invoice within limits established by its owner.
Visa, Mastercard and Ant International are now working on a shared framework for identifying and verifying AI agents involved in payments. The initiative addresses a fundamental problem: payment systems were designed to authenticate people and merchants, not autonomous software.
For Africa, this is not a distant technology debate. It is a strategic question about who will define the next generation of digital commerce.
From human approval to delegated authority
Today’s payment systems typically assume that a human is present at the decisive moment.
A mobile-money user enters a PIN. A cardholder approves a transaction. A bank checks the identity, device and behaviour associated with the account.
Agentic commerce changes that model.
The customer may authorize an AI agent once and allow it to make multiple decisions within a defined mandate. The instruction could be narrow, such as buying an airline ticket below a certain price. It could also be continuous, such as managing recurring business purchases within a monthly budget.
This moves the central payment question from “Is this the customer?” to “Did the customer authorize this agent to make this specific decision?”
That is a far more complex problem.
A trusted system must be able to identify the agent, verify its owner, understand the limits of its authority and create an auditable record of why a transaction occurred.
Without that infrastructure, merchants and financial institutions cannot easily distinguish a legitimate AI purchase from fraud, account takeover or automated abuse.
Africa enters from a position of strength
Africa is not starting from zero.
Mobile money has already normalized digital transactions for consumers and small businesses that were underserved by traditional banking. It has also created infrastructure for instant transfers, merchant payments, savings, credit and cross-border services.
That experience matters.
African payment systems have solved difficult problems around low-cost transactions, basic mobile devices, agent networks and users with limited access to conventional financial services. Mobile-money operators also hold rich information about transaction patterns and account behaviour that could support new forms of risk assessment.
But the architecture remains centred on direct human authorization.
A user initiates or approves a transaction, usually through a handset linked to a registered account. An AI agent making purchases in the background does not fit neatly into that structure.
If agentic commerce develops mainly through global card networks, Africa risks participating only at the final acceptance layer. Local consumers may use the technology, but African mobile-money providers, fintech companies and regulators could have little influence over the underlying standards.
The real battle is control
The most important question is not whether AI agents will make payments. It is who will control their identity, permissions and access to financial accounts.
For banks and mobile-money operators, AI agents could become a new interface between customers and financial services. Instead of opening several applications, a user might instruct one agent to compare prices, move money and complete a purchase.
The institution controlling that agent could gain significant influence over customer relationships and purchasing decisions.
It could determine which merchants are considered, which payment rail is selected and how financial products are presented. Over time, the AI agent could become more important than the banking or mobile-money application behind it.
This creates a strategic risk for African financial institutions. They could retain the regulated accounts and payment infrastructure while global technology platforms control the customer experience and transaction choices.
The value would shift upwards, from the institution moving the money to the intelligence deciding where the money goes.
Liability remains unresolved
Agentic payments also create a difficult accountability problem.
If an AI agent purchases the wrong product, exceeds its budget or misunderstands a customer’s instruction, who carries the loss?
Possible responsibility could sit with the consumer, the provider of the AI agent, the bank, the mobile-money operator, the merchant or the payment network. Each participant may argue that another part of the chain made the critical error.
The problem becomes more serious when an agent’s decision is technically authorized but commercially unreasonable.
A transaction may pass every identity and security check while still failing to reflect what the customer intended. Traditional fraud controls are not designed to resolve that distinction.
This means agentic commerce will require more than stronger authentication. It will need clear rules covering consent, spending limits, dispute resolution, data access, revocation and liability.
Those rules should be established before the technology reaches mass adoption, not after consumers begin suffering losses.
Regulation must evolve without freezing innovation
African regulators face a familiar challenge.
Moving too slowly could allow private global standards to become entrenched before local priorities are considered. Moving too aggressively could prevent African financial institutions from experimenting with a technology that may reduce costs and expand access.
The better approach is controlled participation.
Central banks and financial regulators could establish supervised pilot environments in which banks, mobile-money operators and fintech companies test delegated payments under clear limits.
These pilots should answer practical questions:
- How does a user authorize an AI agent?
- How narrowly can its permissions be defined?
- Can authorization be revoked instantly?
- What information must be disclosed before a purchase?
- Who compensates the customer when the agent makes an error?
- Can an agent choose between mobile money, bank transfers and cards?
- How are suspicious automated transactions detected?
Interoperability will be particularly important. Africa should avoid creating a system in which agentic payments work only for internationally connected cardholders while excluding the mobile-money users who form the foundation of its digital economy.
The Baobab View
Africa’s mobile-money revolution succeeded because the continent developed payment systems around its own market realities.
Agentic commerce presents a similar opportunity, but the window to shape it may be narrower.
Visa, Mastercard and Ant International are already working on the trust infrastructure required for AI-led transactions. African banks, telecommunications companies, fintech firms and regulators should not wait for those standards to arrive as finished products.
They should help define how an AI agent proves its authority, how consumers retain control and how liability is distributed when automation fails.
The next phase of payments will not be defined only by who moves money fastest or cheapest. It will be defined by who controls the machine making the decision.
Africa led the transition from cash to mobile money. Whether it leads the transition from human-initiated to machine-initiated commerce will depend on whether it acts early enough to shape the rules.
