Agentic commerce is about to reinvent payments

There are moments in technology when something shifts so fundamentally that the existing infrastructure suddenly looks like it belongs to another era. We saw it with the internet, when pages became platforms. We saw it with smartphones, when every business suddenly needed an app. Now we’re seeing it again as artificial intelligence moves from answering questions to taking actions.

The latest report from Visa and Artemis Analytics argues that we have entered the age of agentic commerce, where AI doesn’t just recommend products or compare prices but discovers services, evaluates them, negotiates terms and completes transactions autonomously.

This is a much bigger change than many people realise.

For decades, payment systems have been designed around humans. We browse websites, click checkout buttons, authenticate ourselves and approve purchases. AI agents don’t work that way. They discover an API, evaluate whether paying a few cents for data or compute power is worthwhile, make the payment in milliseconds and move on to the next task. They may perform hundreds or thousands of these transactions in the time it takes us to order a coffee.

That’s why the report makes an important distinction between agent-assisted shopping and machine-native commerce.

The first is familiar. Your AI books your flights, renews subscriptions or orders office supplies on your behalf. These are essentially today’s ecommerce transactions with a software agent replacing the human at checkout.

The second is far more revolutionary. Here, AI agents buy API calls, datasets, GPU time, software functions and digital services worth fractions of a cent, often from suppliers they have never encountered before. The payment becomes inseparable from the work itself.

This changes everything.

The internet was never really built for micropayments. HTTP even reserved error code 402 Payment Required back in the 1990s, anticipating a future where machines could pay one another, but card economics made such transactions impractical. Processing fees were simply too high. Advertising became the internet’s business model because charging pennies wasn’t economically viable.

AI has changed that equation.

The arrival of capable reasoning models such as Claude and GPT means there is now genuine demand for autonomous, high-frequency purchasing. At the same time, blockchain infrastructure has evolved to deliver sub-cent settlement costs and near-instant finality, making these tiny transactions commercially viable for the first time. According to the report, demand came first and infrastructure followed.

This is where things become particularly interesting for banks and payment providers.

The report highlights two emerging protocols, x402 and the Machine Payments Protocol (MPP), designed specifically for machine-to-machine payments. They sit above existing payment rails rather than replacing them, enabling APIs, developers and data providers to sell services directly to software agents without requiring user accounts or subscription relationships.

This means that money is no longer something humans own and spend. Money increasingly thinks, decides and acts. Once agents have delegated budgets, identities and authority, they become economic actors in their own right.

That raises entirely new questions.

Who is responsible if an AI buys the wrong service? What happens if one compromised agent makes ten thousand fraudulent purchases before anyone notices? How do disputes work when one AI has paid another AI, which has subcontracted work to five more AIs? Existing legal frameworks were never designed for chains of autonomous delegation.

This is why trust becomes the defining challenge.

The payment itself is almost trivial. Identity, reputation, delegated authority, programmable spending limits, liability and governance become the real infrastructure. That’s why we’re seeing Visa, Mastercard, Stripe, Google, Coinbase and others building frameworks for agent identities, tokenized credentials and programmable controls rather than simply faster payment mechanisms.

Perhaps the most significant conclusion from the report is that this isn’t shaping up as a battle between traditional finance and crypto. Instead, both worlds appear to be converging. Cards remain well suited for larger delegated purchases while stablecoins and blockchain rails increasingly power high-frequency machine commerce. The future looks hybrid rather than ideological.

The mistake would be to dismiss this as another payments innovation because what we’re witnessing is the emergence of a new economic layer for the internet. The last thirty years connected people to websites. The next thirty years will connect autonomous software to autonomous software. In other words, bot-to-bot commercial economies. When billions of AI agents begin buying billions of digital services every second of every day, commerce stops being something humans initiate and becomes something machines continuously optimise.

That is not just a new payment system. It is a new economy.

You can read the full Visa report here or view below: