When Stripe launched Link's wallet for agents on 29 April 2026, the launch post was plain about approval: "Today, each request requires the person's review before the credential is shared with your agent." Link's support pages now describe a second mode: "You can set up a Link balance to let your agent make purchases without requesting your approval each time."

So the choice every agent builder faces is now a setting in a wallet that Stripe's launch post credits with "more than 200 million consumers." You pick a number, and under it the agent stops asking.

How does approval work in Link's agent wallet?

Link's wallet for agents runs on spend requests. Your agent asks to spend a specific amount at a specific seller, and Stripe's docs describe the next step in one line: "The customer approves it on the Link website or in the mobile app." Link then returns a payment credential for that purchase. Stripe's docs give the customer 30 minutes to respond before the request expires.

The April launch post says the agent "can create a spend request to get either a one-time-use card or an SPT", meaning a Shared Payment Token. The request carries a written explanation for the person approving. Stripe's docs require at least 100 characters of it and tell builders to name the item, the seller and why the agent is buying.

Link's support FAQ gives two windows. One answer says "You have 30 minutes to approve or deny the spend request", another says "The spend request expires after 10 minutes." Plan for the shorter one.

Two pages give two answers on who can use it. The docs say "Agent payments are available to US and Canadian customers." Link's support FAQ lists five countries: "The Link agent wallet is available to users who created their accounts in the US, Canada, Mexico, Japan, and Singapore."

What do Link's support pages say about skipping approval?

Link's support pages now describe a balance you fund in advance, with a limit under which the agent buys without asking. Link's balance FAQ offers it to customers "using Link for agentic purchases in the US". The setup step reads: "Set a purchase limit. This is the maximum amount your agent can spend per purchase without requesting your approval." Purchases over the limit still go to the customer.

Setup takes identity verification, with the customer's date of birth and the last four digits of their Social Security number, and money in the balance is held as USDC. The off switch is one sentence: "To prevent your agent from skipping approvals, set the limit to $0." Stripe has published no dated announcement of the feature that we could find.

The setup page describes one limit the customer chooses, per purchase. Totals exist, but Link sets them. Its FAQ says: "Your spending limit may vary depending on your purchase behavior and the agent you're using. If you reach your limit, it resets daily. It can't be increased."

Stripe's docs also describe a user-info response that includes agent_wallet_spend_limits, "with per_transaction, daily, and thirty_day entries carrying limit, used, and remaining." Read those values before you assume how many under-limit purchases your agent can make in a row.

A budget the customer sets is still ahead. Stripe's 29 September 2026 post says: "We plan to introduce spending controls that let consumers set an agent's budget".

When is approving every purchase the right setting?

Approving every purchase is the right setting for an agent that buys rarely, or buys something different each time. Each request then carries new information: a seller the customer hasn't seen or a price nobody could predict. Reading one request a week costs your customer a minute.

It stops working when the requests repeat. An agent that reorders the same 15-euro item four times a week sends your customer four identical screens. Sooner or later they approve without reading, and the approval has become a tap that proves nothing. That's the case a threshold is for.

How do you choose the approval threshold?

Set the approval threshold just above the largest purchase your agent makes routinely, and below the smallest purchase your customer would want to read first. List what the agent bought over the last month, find the price where routine purchases stop and unusual ones start, and put the line in that gap. With no clear gap, keep approving everything until one appears.

Say your agent keeps a small office supplied. Its orders run 12, 18 and 30 euros, and once a quarter it buys something large, like a 240-euro chair. The gap sits between 30 and 240. A threshold of 35 euros lets every routine order through and sends the chair to your customer.

Now stress the number. A per-purchase threshold of 35 euros says nothing about ten 30-euro orders in one night, which is 300 euros with no request sent. That's why a threshold needs a total beside it, whether that's a prepaid balance kept small or a monthly budget.

Price changes are the other edge. Stripe's September post says "Link now supports incremental authorization, which lets agents request a higher approved amount when the final price exceeds the original." Shipping and tax can push a 33-euro basket to 38. Ask any provider whether an increase that crosses the threshold goes back to your customer.

A threshold also can't see what kind of purchase is under it. A 9-euro subscription that renews every month fits under any sensible limit and deserves a look anyway. When to require a human approval covers the purchases worth reading at any price.

Where does Shatale put the line?

Shatale puts the line in two numbers your customer sets in the Shatale app: a monthly budget and a per-purchase cap. Inside those, your agent pays on its own, unless the service asks for their approval or the price moved before the charge. Over them, the purchase is blocked or sent to your customer to approve or decline, with the merchant, the amount and the agent in front of them.

Every purchase runs on a Shatale card assigned to that purchase alone and capped at its amount. Each decision lands in a tamper-evident audit log, per agent, so the purchases your customer didn't review are still there to read afterwards. Shatale does card purchases only.

FAQ

Can an agent using Stripe Link buy without the customer's approval?

Yes, in the US, under a limit the customer sets. Link's support pages describe a Link balance that lets "your agent make purchases without requesting your approval each time", up to a maximum amount per purchase that the customer chooses during setup. A purchase above that amount still needs their approval. Without the balance, every spend request waits for the customer.

How does a customer stop a Link agent from skipping approvals?

They set the per-purchase limit on their Link balance to zero. Link's support page says: "To prevent your agent from skipping approvals, set the limit to $0." To change the limit they ask the agent for a link to the settings page, and with the limit at zero every purchase goes to them for approval again.

Who can use Link's wallet for agents?

Stripe's pages disagree. Its developer documentation says "Agent payments are available to US and Canadian customers." Link's support FAQ says the agent wallet "is available to users who created their accounts in the US, Canada, Mexico, Japan, and Singapore." The balance that skips approval is narrower. Link's balance FAQ describes it for agentic purchases in the US, and setup needs the last four digits of a Social Security number.


When to require a human approval goes through the purchases a threshold shouldn't wave past.

Shatale is the control layer for AI-agent payments. Its authorization architecture is the subject of European patent application EP26194994.5 (filed; priority 28 July 2026).