How autonomous software agents may reshape procurement, payments, and digital commerce.
Artificial intelligence is starting to change more than how we search for information. It may also change how economic decisions are made.
For most of the internet era, software has helped people discover, compare, and purchase products. But the final step of the transaction has almost always been human.
That model may be starting to shift.
A concept often referred to as agentic commerce describes a world where AI systems can discover products, evaluate options, and execute transactions on behalf of people or businesses.
Recent developments from major AI labs suggest this idea is moving quickly from theory to practice. OpenAI’s Operator agent was introduced as a system that can use a browser to complete tasks such as filling out forms or placing online orders. Anthropic has introduced similar capabilities through Claude’s computer-use tooling, which allows AI to interact with desktop environments using mouse and keyboard controls.
For payments companies, merchants, and software platforms, this raises an important question:
What happens when software, not people, starts making purchasing decisions?
A simple way to think about agentic commerce is this:
Humans define intent. AI agents evaluate options. Infrastructure executes the transaction.
Let’s explore what agentic commerce is, why it is emerging now, and what it may mean for payments infrastructure.
Agentic commerce refers to systems where AI agents perform economic actions on behalf of users.
Instead of navigating websites, comparing suppliers, and completing checkout flows manually, a person defines an objective and allows software to carry out the task.
For example, someone might instruct an agent:
Buy the cheapest standing desk under $600 that can be delivered this week.
The agent could then:
The person defines the goal. The software handles the process.
This moves commerce away from interface-led decision making and toward automated decision systems.
Three developments are converging to make this possible.
AI models are no longer limited to answering questions. They can interpret goals and reason through multi-step tasks.
Systems like OpenAI’s Operator and Anthropic’s computer-use capability demonstrate how AI agents can navigate software environments and perform real-world tasks.
That matters because purchasing is rarely a single step. Even simple transactions involve searching, filtering, comparing, and confirming.
The second shift is orchestration.
Agent systems can combine reasoning with tools, interact with external services, and move through multi-step workflows. Instead of answering a prompt and stopping, they can continue executing tasks until an objective is reached.
This capability allows AI to transition from analysis to action.
Much of modern commerce already runs through APIs and automated systems.
Subscriptions, procurement workflows, and recurring billing are often triggered by software rather than manual transactions. For example, Stripe’s subscription infrastructure manages recurring payment lifecycles programmatically through APIs.
Agentic commerce builds on top of this existing automation layer rather than replacing it.
As this ecosystem evolves, it can be useful to think about three categories of agents.
Demand agents represent buyers.
They may act on behalf of individuals, households, finance teams, or procurement departments.
Examples include:
Their role is to identify the best option within defined constraints.
Supply-side systems represent merchants or vendors.
These systems may control:
Rather than presenting static listings, merchant infrastructure may increasingly respond dynamically to demand signals.
The final layer handles execution.
These systems manage:
This is the layer where payments infrastructure becomes essential.
One way to understand this ecosystem is as a layered system.
This is where a person or business defines a goal.
Examples include:
Humans define the objective.
AI agents interpret the goal and evaluate options.
They may compare:
This is the layer changing most rapidly.
Suppliers respond to incoming demand.
Merchants may expose machine-readable information such as:
Over time these signals may be consumed directly by AI systems.
At the base sits the infrastructure that executes transactions.
Even if AI makes the decision, this layer still moves the money.
This means payments infrastructure may become more important, not less.
This shift is already visible in the payments industry.
Visa launched Visa Intelligent Commerce to enable AI agents to search for and purchase products through its payment network. Mastercard introduced Agent Pay as a framework for enabling secure transactions initiated by AI systems.
These initiatives suggest that payment networks expect AI-driven transactions to become commercially meaningful.
They also highlight key infrastructure questions:
These are not interface questions. They are payments and governance questions.
Today’s digital commerce is still built around human interfaces.
Websites, apps, and checkout flows are designed for people.
AI agents interact differently. They rely on structured data, APIs, and automated workflows.
The direction of travel suggests competitive advantage may increasingly depend on:
Interface design will still matter, but machines may increasingly evaluate the options.
Agentic commerce is unlikely to transform all retail at once.
The earliest adoption will likely appear in environments where purchasing decisions are already structured and repeatable.
Examples include:
These environments already rely heavily on software systems and decision rules. AI agents simply add more flexible reasoning on top of existing automation.
In the near term, AI systems will likely assist with research and comparison while humans complete the final purchase.
The next phase may involve agents executing transactions inside defined governance frameworks.
These could include:
Many procurement systems already operate this way today.
In a more advanced model, agents may operate within broader delegated authority.
Examples could include:
Even in this scenario, payments infrastructure remains essential.
Some industries may encounter agent-driven transactions sooner than others.
Financial professionals may increasingly design governance frameworks for automated spending and ensure clean reconciliation of software-initiated transactions.
Service businesses may find procurement agents evaluating vendors based on reliability, pricing transparency, and billing predictability.
Supply chains already rely heavily on structured purchasing workflows. Inventory systems may eventually trigger automated purchasing decisions.
AI agents may help businesses compare software tools, manage subscriptions, and optimise software spending.
At Pinch we spend a lot of time thinking about how payments infrastructure evolves as commerce changes.
Historically many payment systems were designed around human interaction. Someone opened an invoice, clicked a checkout page, or entered card details.
Agentic commerce introduces a different possibility.
Transactions may increasingly originate from software systems acting on behalf of people or businesses.
In that environment payments infrastructure must support more than checkout flows. It must support programmable transactions, governance controls, reliable APIs, and clean reconciliation with financial systems.
As software agents become more capable, the systems that move money may increasingly act as the trusted execution layer of automated commerce.
Agentic commerce represents a shift in where commercial decision making happens.
Humans define intent. Software interprets the goal and moves through the workflow. Infrastructure executes the transaction.
If more economic activity is initiated by software, the systems that authorise, route, settle, and reconcile payments become even more important.
Recent developments from AI companies, payment networks, and enterprise software providers all point in the same direction.
More commerce is becoming software-mediated.
And the infrastructure that moves money will remain central to how that system works.