AI Agents That Pay: Setting Limits Before You Start

AI compares the quotes, places the order and settles the payment — and a transaction has now been completed that way from start to finish, with no human intervening midway. In this module we use that event as our subject and work through, in concrete steps, how a Japanese company operating in the Philippines should define the boundary when it hands work to an AI agent.

According to published reports, Lianlian DigiTech used its AI agent LoopXPay to complete the first live business-to-business agentic transaction in Greater China. The transaction was supported by Visa's Agentic Directory, which underpins trusted interactions between AI agents. The agent identified the purchasing requirement, recommended suitable suppliers, compared the options, placed the order and securely executed the payment, all within a single workflow. What matters is that the whole sequence ran within pre-defined spending controls and approval parameters. The announcement was made in July 2026 and is positioned as showing how purchasing and payments can be streamlined for small and medium businesses. In this module we work through the implications in Parts 1 to 4, applying them to your own situation.

Part 1: Read, Then Consider the Implications for Your Company

Three points stand out in the reporting.

Point	Detail
What moves has changed	Not text or data, but the company's actual money
The boundary was set first	It operated within pre-defined spending limits and approval conditions
The target is everyday small orders	Framed as easing procurement for small and medium businesses, not large-scale sourcing

Step 1: Pre-Reading (3 min)

Before you read, consider how this applies to your own company.

Which of your purchasing tasks are small in value and repeated every month?
Up to what amount may those purchases be made without anyone's approval?
Is that rule written down, or does it live in one person's head?

Step 2: First Reading (10 min)

The following is a fictional internal memo, written for this module with a Japanese company in the Philippines as the subject, based on the facts of the news.

Internal memo: A transaction where AI executed the payment, and what we should decide first

Reports indicate that a business-to-business transaction has been completed in which an AI agent handled everything from selecting the supplier to executing the payment, in a single workflow. Three points deserve attention.

First, what AI moves has shifted from information to money. The AI we have used internally so far has mainly written text and organised data. In this case, company funds actually moved. That is a different kind of risk, because it is harder to undo when something goes wrong.

Second, the boundary was set before the agent was allowed to run. Reports state that it operated within pre-defined spending controls and approval conditions. In other words, the sequence was not "it is clever, so we trusted it" but "we defined what it could be trusted with, so we could trust it."

Third, it is being discussed as a way to lighten purchasing for smaller businesses. The subject is not large-scale corporate procurement but the small, routine orders placed day to day. The purchasing our Philippine site does is exactly that size.

Our three takeaways: (1) set the ceiling for how much may be ordered automatically, (2) write out whose approval is required at each value band, and (3) trial it on a single item first.

Source: Visa and Lianlian Advance Trusted B2B Agentic Commerce Through LoopXPay's First Live B2B Agentic Transaction — PR Newswire (July 2026)

Note: the business scenario above is a fictional internal memo written for learning purposes on the basis of published reporting. Systems and service availability can change, so please check the primary source linked above for current detail.

Step 3: Comprehension Check (5 min)

What steps did the AI agent carry out, in order, during this transaction?
What defined the boundary the agent operated within?
What size of company is this presented as an efficiency gain for?

Step 4: Three-Minute Briefing (10 min)

Practise explaining this news to your management team in three minutes. Presenting it in the order "what happened (AI completed a purchase and the payment without human operation) → why it matters (what AI moves has shifted from information to money) → what we should do (document the spending ceiling and approval conditions first)" tends to land well.

Part 2: Key Terms Explained (for Management)

Agentic transaction: a transaction in which AI proceeds by its own judgement rather than waiting for instructions at each step. Earlier AI stopped at making a recommendation, and a person carried out the action.

Spending ceiling: the maximum an AI may spend in one transaction, or within a given period. Without this, the scope of what you are delegating cannot be defined.

Approval conditions: the rule setting out which amounts require human sign-off. It only works paired with the ceiling.

Agent identity verification: the mechanism for confirming that the counterparty is a legitimate agent. It plays the same role as checking who a company is before dealing with them.

Part 3: Applying This to Your Company

Decide these in order before you begin.

Order	What to decide	If you do not
1	The ceiling below which ordering may be automatic	It runs with the boundary left vague
2	Whose approval is needed at each amount band	Nobody is in a position to stop it
3	Limit the first trial to a single item	You cannot identify the cause when something goes wrong

Sort purchasing into three value bands

Start by sorting your Philippine site's purchasing by value. Daily consumables, monthly recurring orders, and everything larger is enough of a split. Only the bottom band is a candidate for delegation to AI.

Put the ceiling and the approvals on one page

Write three lines on a single sheet: "up to X pesos per transaction", "up to X pesos per month", and "above that, whose approval". You do not need a long policy document. Work on the basis that anything not written down cannot be operated.

Trial one item for two to four weeks

Do not extend this to all purchasing at once. Pick a single item — office supplies, cleaning supplies, whatever suits — and run it. In that period you will always find at least one behaviour you had not anticipated.

Part 4: Common Failure Patterns (What Not to Do)

Failure 1: Trialling it without setting a ceiling, to "see how it goes"

If you run a trial without a ceiling, then when something goes wrong you cannot tell what was inside your expectations and what was outside them. The spending ceiling is not there to test capability. It is there to decide the size of your failures.

Failure 2: Naming an approver by job title and ignoring absence

If you decide "the department head approves", the process stops while that person is travelling. When approvals stall, the team quietly reverts to the old way of working. Name a deputy approver as well.

Failure 3: Delegating without keeping records

If you cannot later trace when, what, at what price and on what basis something was ordered, your accounting team cannot verify it. Make sure the same information is retained as when a person places the order, before you extend the scope.

Failure 4: Extending it without checking local commercial practice

In the Philippines, relationships with suppliers and payment terms are sometimes settled verbally rather than on paper. A system that automatically switches suppliers can cut across an existing relationship, so check with your local staff first.

Three Tips for Getting Value from This

Write the ceiling before anything else: decide how much you are delegating before you evaluate the technology. Once that is fixed, the remaining decisions get faster.
Start with a small item: trialling at a scale where failure is recoverable is, in the end, the quickest route to wider use.
Go to primary sources: service scope and terms will keep moving. Check the provider's own announcements rather than summary articles.

Bonus: How to Make Use of PH AI Works

Working out which of your purchasing tasks can safely be delegated, and how to word the ceiling and approval conditions, tends to stall when handled internally. In a free PH AI Works consultation we take questions from Japanese companies operating in the Philippines on separating out purchasing tasks and getting the delegated scope onto a single page. Accounting and tax judgements themselves belong to qualified professionals; our part is helping you build the internal workflow.

Sources

Visa and Lianlian Advance Trusted B2B Agentic Commerce Through LoopXPay's First Live B2B Agentic Transaction — PR Newswire (July 2026)
Visa and Lianlian complete Greater China's first live agentic B2B payment — Electronic Payments International
Visa, LianLian Automate Cross-Border B2B Payment With AI Agent — Caixin Global (27 July 2026)
Visa, Lianlian complete first B2B AI agent payment — CFOtech Asia