From shopping cart to mandate: how payments made by agents will work
Agentic payments need consent, limits, traceability, and trust. We explain mandates, confirmations, and risks for companies.
Online shopping has always assumed something basic: a person looks at a screen and clicks "buy." AI agents break that assumption. If an agent can search, compare, and prepare a purchase, the system needs to prove it is acting with real user permission.
That is where a key idea appears: the mandate.
What a mandate is in agentic commerce
A mandate is a verifiable instruction that defines what an agent can do. For example:
- Buy white running shoes for less than 120 euros.
- Book a hotel with a maximum budget of 700 euros.
- Reorder office supplies if the price stays under a limit.
- Buy tickets only if they appear in a specific time window.
The difference from a vague order is that the mandate sets conditions: price, product, time, provider, limits, and approval.
Why "the user asked for it" is not enough
In traditional payments, the merchant sees the user on its website. In agentic commerce, it may receive a request initiated by an agent. The merchant, bank, and user need to know:
- Who authorized the action.
- What exactly was authorized.
- What limits the agent had.
- What was shown to the user.
- What was confirmed before payment.
- Who is accountable if there is an error or fraud.
Without that chain of evidence, agentic commerce cannot scale with trust.
Human confirmation and limited autonomy
Not all payments need the same level of control. An agent can prepare a purchase while final confirmation remains human. In other cases, the user may delegate a purchase with very clear limits.
Example: "buy printer ink when it drops below 40 euros and is compatible with this model." The agent does not decide freely. It executes an instruction within predefined conditions.
What merchants should prepare
Companies that sell online should start thinking about:
- Clear, structured product information.
- Return policies that AI can read.
- Signals to distinguish legitimate agents from malicious bots.
- Clear confirmations and receipts.
- Systems that preserve the customer relationship.
- Records of consent and conditions.
The checkout of the future will not only be a page. It will be a conversation between user, agent, merchant, and payment network.
Risks for SMEs
SMEs may be left out if their systems are not readable or trustworthy for agents. They may also take on risk if they accept automation without verifying identity, intent, and authorization.
The rule is simple: automate, but with proof of consent and limits.
How this relates to Polp
Although Polp does not process payments, the logic is the same as in company knowledge: important actions need sources, permissions, and evidence.
In agentic commerce, trust will not come from an agent that "seems useful." It will come from proving which mandate it followed and which user approved it.
For a B2B SaaS like Polp, this shift also points to an SEO direction: content must be clear, structured, and easy to interpret for both people and AI agents.
Sources:
Stop searching. Start asking.
Upload your PDFs, spreadsheets, and docs. AI handles the rest.
Get started