Agentic commerce is shopping where an AI agent, acting on your instructions, searches for products, compares options and completes the purchase for you. Payment networks are building it around tokenised card credentials, spending limits you set, and confirmation steps so the agent can only spend what you allow.
The idea has moved quickly from concept to pilot programmes. Below we explain how it works, where the risks are, and what consumers and businesses can do to prepare.
What does “agentic” mean in AI?
A chatbot answers questions. An AI agent goes further: it can plan a series of steps and take actions, such as searching several stores, filling in a basket, booking a slot or paying an invoice. “Agentic” simply describes software that acts with some autonomy towards a goal you gave it.
In commerce, that means you might say “find me a pair of waterproof running shoes under a set budget and order them in my size”, and the agent handles the search, the comparison and the checkout.
How do AI agents pay for things?
The hard part is not finding products; it is paying safely. Typing a full card number into an AI tool would be risky. The card networks’ answer is tokenisation: replacing your real card number with a digital token that can be limited, tracked and switched off.
What have the card networks announced?
- Visa Intelligent Commerce was announced by Visa in April 2025. Visa describes “AI-ready cards” that replace card details with tokenised credentials, confirm that an agent is allowed to act for the consumer, and let consumers set spending limits and conditions.
- Mastercard Agent Pay was announced by Mastercard on 29 April 2025. It introduces “Agentic Tokens” that build on the tokenisation already used for mobile contactless payments and saved cards.
Both approaches share the same principles: the agent never sees your real card number, you decide the rules, and the network can apply its usual fraud checks to each transaction.
What does a typical agent purchase look like?
- You link a payment method to an AI assistant and approve it, usually with your bank or card issuer’s authentication.
- You set limits: a maximum amount, approved merchants or categories, or a time window.
- You give the agent a task in plain language.
- The agent researches options and, depending on your settings, either buys within your limits or asks you to confirm.
- The merchant receives a token instead of your card number, and you see the transaction in your banking app as usual.
What are the benefits of agentic commerce?
- Time saved: repetitive purchases such as groceries, refills and travel bookings can be delegated.
- Better comparisons: an agent can check more options than most people have patience for.
- Business automation: companies can let agents handle routine procurement, supplier payments or subscription renewals within strict policies.
- Security by design: tokens can be restricted to one agent or merchant and revoked without replacing your physical card.
What are the risks of letting AI shop for you?
Delegation brings new questions that are still being worked out by networks, banks, merchants and regulators.
Mistakes and misunderstandings
An agent can misread an instruction, choose the wrong size or book the wrong date. Check how returns, cancellations and disputes work when an agent made the purchase, and keep confirmation steps on for anything expensive.
Fraud and manipulation
Fraudsters may try to trick agents with fake shops or hidden instructions on web pages. This is why verified agents, tokenised credentials and spending controls matter, and why you should only connect payment methods to AI services you trust.
Privacy
Agents work better with more information about your preferences. Read what data an assistant collects, whether you can limit it, and how to delete it.
How can businesses prepare for AI shopping agents?
If agents become a common way to shop, they will “read” your store differently from people. A few practical steps help either way:
- Publish clear, structured product data: accurate titles, prices, sizes, stock levels and specifications.
- Keep policies on delivery, returns and refunds easy to find and unambiguous.
- Talk to your payment provider about tokenised payments and how they plan to support agent-initiated transactions.
- Review fraud rules so legitimate agent traffic is not blocked while suspicious automated activity still is.
- Make sure your site loads quickly and works without complex steps that confuse automated visitors.
Checklist: before you let an AI agent pay for you
- Use only reputable AI assistants and official payment integrations.
- Set a spending limit and, where possible, restrict merchants or categories.
- Keep “confirm before buying” switched on for large or unusual purchases.
- Turn on transaction alerts in your banking app.
- Know how to revoke the agent’s payment access quickly.
- Review the agent’s purchase history regularly.
For more practical explainers on technology and business, browse the Codelude news section.
Frequently asked questions
Is agentic commerce available today?
Visa and Mastercard announced their agent payment programmes in April 2025 and have been working with AI and payments partners. Availability depends on your country, bank and the AI service you use, so check with your provider.
Does the AI agent see my card number?
In the network programmes described above, agents use tokenised credentials instead of your real card number, which limits the damage if a token is misused.
Who is responsible if an agent buys the wrong thing?
That depends on the merchant’s policies, your card issuer’s dispute rules and local consumer law. Keep confirmations on and check terms before relying on an agent for important purchases.
Can businesses use AI agents for payments?
Yes. Mastercard’s announcement describes business uses such as sourcing and paying international suppliers with virtual corporate card tokens. Companies should set clear approval policies first.
This article is for general information only and is not financial or legal advice.
At Codelude, we build software products where AI, payments and everyday business meet.




3 Comments