Agentic Commerce and Gift Cards: What Changes for Rewards Programs

Agentic commerce gift cards are gift cards that an AI agent picks, buys, sends or pays with on someone’s behalf. Agentic commerce means the AI agent completes the purchase instead of only recommending it, and gift cards keep coming up because a fixed balance gives the agent a spending limit that’s built in.

TL;DR

• Agentic commerce is buying done by an AI agent: it finds the product, fills the cart and pays, usually within limits a person set.
• Gift cards suit AI shopping agents because the balance caps what the agent can spend, a brand card only works at that brand, and the agent never needs your main company card (where the agent platform accepts gift cards).
• Agentic commerce is real but small in late 2026: one modeled estimate puts agent-completed orders at 0.68% of US ecommerce orders in Q3 2026, and most card-network programs are still in pilots or early rollout.
• For a rewards program, an agent that sends gift cards after a survey or a closed deal needs four controls first: a spend cap, an approval step, a log of every action and a check on who receives the reward.

A few summers ago, on a cabin weekend with friends, I handed one of them my credit card for the grocery run because I was busy losing a board game. She came back with everything on the list, plus a watermelon, three kinds of crackers and a bag of ice we had nowhere to keep. She held up the receipt like evidence: “You didn’t say how much.”

I hadn’t. That was the whole problem.

An AI shopping agent raises the same question for anyone who runs a rewards program. You can hand it the task. You also have to hand it a limit.

That’s where gift cards come in, and it’s why they keep turning up in payments research and card-network announcements this fall.

What is agentic commerce, in plain terms?

Agentic commerce is shopping where an AI agent acts for a person or a business and finishes the purchase, not just the research. You give the agent a goal (“restock the office coffee under $200” or “send each new participant their thank-you reward”), and it searches, compares, builds the order and pays within the rules you set.

The difference is the last step. A chatbot that suggests three coffee makers is helping you shop. An agent that buys one is doing the shopping.

That last step is what the payment industry has been building for since spring 2025. Visa introduced Visa Intelligent Commerce on April 30, 2025, and said: “Each consumer sets the limits.” Mastercard unveiled Agent Pay a day earlier, built on “agentic tokens” that stand in for the card number. OpenAI and Stripe published the Agentic Commerce Protocol on September 29, 2025, and Google announced the Agent Payments Protocol (AP2) on September 16, 2025.

Why do gift cards keep coming up in agentic commerce?

Gift cards keep coming up because they already behave the way the industry wants an agent’s payment to behave: a fixed amount, a clear scope and no link to your main account. A card worth $50 can’t be talked into spending $500.

Three properties do most of the work:

• A set budget. The balance is the ceiling. An agent paying with a $50 gift card can’t overspend it, whatever it misreads in a prompt.
• Brand-scoped spend. A brand gift card only works at that brand. If an agent’s job is “buy the team lunch from our usual place,” a card for that place keeps the purchase there.
• No stored primary card. The agent holds a card worth a known amount, not your company credit card. If something goes wrong, the most you can lose is the balance.

The networks and AI platforms are rebuilding the same ideas for cards. In OpenAI’s delegated payment flow, it “sets a maximum chargeable amount and expiry” for a one-time payment. Google’s AP2 has users sign an “Intent Mandate” with “price limits, timing, and other conditions” before an agent may buy on its own. The gift card had a hard limit before anyone called it a mandate.

The honest catch is acceptance. Whether an agent can pay with a gift card depends on the agent platform and the merchant, and few of them say much about it yet. So read “gift cards for agentic commerce” as “where supported” for now.

The people who buy gift cards are paying attention anyway. In the AI & Branded Payments Study, Blackhawk Network’s online survey of 694 US leaders responsible for incentives, commerce and branded payments (April to June 2026, released October 19, 2026), 24% picked AI assistants that “purchase and deliver the right gift card for preapproved occasions” as the most likely way AI will reshape gift card buying by 2030.

How will AI agents change buying and sending rewards?

AI agents change rewards programs in two places: how recipients spend a reward, and how senders pick, buy and send one. The second is the part you control, and it’s usable today.

On the recipient side, the change is mostly ahead of us. The same study’s release describes an assistant that could “surface an existing gift card when a consumer is shopping at a relevant retailer,” so a forgotten reward gets remembered at the right checkout.

On the sender side, the change is a smarter version of the automation many teams already run. Today a typical setup is a rule: when a survey response lands, send a $25 card. An agent adds judgment on top of that rule. It can check who qualified, prepare the order and ask you to confirm it, and it can answer “did everyone in last week’s study get paid?” without anyone opening a spreadsheet.

Here’s how that looks for different teams:

Take a research team paying 200 participants $25 each after a study: $5,000 leaves the account. Today someone exports a list, checks it against the session log and places the order. With an agent, the checking and drafting happen in a conversation, and a person still approves the $5,000. The work shrinks. The responsibility doesn’t.

If you already send rewards from your own product, our guide to automating gift card rewards through an API after a user action covers the rule-based version, and the Gift Card API guide covers the wider setup.

What controls does a rewards program owner need before an agent spends?

A rewards program owner needs four controls before any agent sends a reward: a spend cap, an approval step, a log of every action and a check on who gets paid. None of them depend on how well someone wrote a prompt.

1. Spend caps. Set a limit per order, per day and per month, separate from the account balance. A prefunded balance is a cap too: the agent can’t spend money that isn’t there.
2. Approvals. Decide the amount above which a person confirms. Start with “every order” and loosen it once the logs look boring.
3. Logs. Every action the agent takes (a lookup, a draft, a send, a resend) should land in a record you can read without asking the agent what it did.
4. Recipient verification. The agent should only send to recipients your own records confirm, and the recipient should confirm who they are when they claim the reward.

The fourth matters more than it looks. Researchers at Palo Alto Networks’ Unit 42 described a hypothetical attack (March 2026) in which hidden text on a deals website tells a shopping agent to add a gift card for an outsider to the cart. If the person approving only sees a total, they approve it. Showing every line and every recipient at approval time closes most of that gap.

Shoppers already want this kind of final say. In First Page Sage’s agentic commerce study (October 2026), 6.8% of US online shoppers had let an agent check out with their approval, against 1.3% who had let one buy with no approval at all.

What’s live today, and what’s still announced?

Agentic commerce is live in pilots, early rollouts and a few in-app checkouts, but it’s small and the plans keep changing.

The adoption numbers say “early.” First Page Sage, a marketing agency, estimates that agent-completed orders rose from 0.27% to 0.68% of US ecommerce orders between Q2 and Q3 2026. It’s a modeled figure, built from a panel of 4,936 US shoppers and order data from 43 brands, so read it as a direction, not a census. Trust is the brake. In a Wildfire Systems survey of 1,007 US consumers (October 20, 2026), only 5% would trust an AI platform itself to process an agent’s payment. Comfort also drops with price: 29% were very comfortable with an agent buying something under $20, and 6% at $500 or more.

Here’s where the main programs stood when we checked their official pages on November 13, 2026:

The OpenAI row is the useful reminder: a flagship checkout was scaled back six months after launch. Build your controls so they don’t depend on one platform’s plans.

Businesses are preparing anyway. In Blackhawk Network’s study, 23% of respondents said they’re building or piloting for agentic commerce and 47% are actively exploring it.

Seven in ten exploring or building means your vendors and finance team will ask about this soon.

How does Giftronaut fit an agent-ready rewards program?

Giftronaut already lets an AI assistant work with your rewards account through MCP (the Model Context Protocol), with versions of those controls built in. Here’s what that looks like:

• AI tools you may already use. Giftronaut MCP works with Claude, Claude Code, OpenCode, Cursor, Codex and other MCP-compatible clients. Setup takes a Giftronaut developer account, a server URL and a browser sign-in, with no integration code.
• Approval before anything goes out. Before anything is charged or sent, the assistant shows a full preview of every recipient, the amount and the total cost, and waits for your confirmation.
• Spend caps. Every account has a daily and monthly limit on gift card spending through the API or MCP, separate from your balance. New accounts start at $1,000 a day and $10,000 a month, and increases are reviewed manually. Your account is also prefunded, so nothing is spent beyond the balance.
• Recipient checks. Each recipient confirms their email address through a one-time verification step before they reach the redemption page.
• A log. MCP tool calls are logged alongside your API activity in the request log dashboard.
• No guessing preferences. A Choice Card lets each recipient pick from 4,300+ brands across 225 countries, so the agent doesn’t have to guess who wants coffee and who wants books. Recipients choose their country when they redeem.

For the setup itself, our post on the AI-powered gifting assistant walks through connecting your AI tool and the first prompts to try. If you’d rather build agents into your own product, the Giftronaut API (currently in beta) has a self-serve sandbox.

Where to start with agentic commerce this quarter

I still hand over my card for the grocery run. It now comes with one sentence: “Up to this much, and text me before anything over twenty dollars.” My friend calls it my spending policy, which is fair.

Your rewards program needs the same sentence before any agent touches it. Start small this quarter:

1. Write the limit down. Pick a per-order cap, a monthly cap and the amount above which a person approves.
2. Start read-only. Let an assistant check balances, recent orders and delivery status before it drafts anything.
3. Pilot one trigger. Choose one repeat reward, such as research participant payments, and have the agent draft while a person approves every order for a month.
4. Read the logs. If they’re boring, raise the approval threshold. If they’re not, you’ve learned something cheaply.

Do those four things and you’ll be ready for whichever agent your team ends up using, including the ones that change their plans next spring.

Frequently Asked Questions

What does agentic commerce mean?

Agentic commerce means an AI agent completes a purchase for a person or business, not just recommends one. The agent searches, compares, builds the cart and pays within limits the person sets.

Can an AI agent buy gift cards?

Yes, where the agent platform and the seller support it. Card networks and AI platforms announced agent payment programs in 2025, and several are now in pilots or early rollout. Acceptance still varies by platform and merchant.

Are gift cards safer than a credit card for AI shopping agents?

A gift card limits how much an agent can spend, because the balance is the ceiling. It also keeps your main card out of the agent’s hands. It doesn’t stop a wrong purchase, so approvals and logs still matter.

How big is agentic commerce in 2026?

Still small. First Page Sage estimates agents completed 0.68% of US ecommerce orders in Q3 2026, up from 0.27% in Q2. The figure is a modeled estimate, so treat it as a direction.

How do I stop an AI agent from overspending a rewards budget?

Set a cap per order, per day and per month, and require approval above a set amount. Keep the account prefunded so spending can’t exceed the balance. Then review the agent’s log every week.

Can an AI assistant send Giftronaut rewards?

Yes. Giftronaut MCP connects Claude and other MCP-compatible AI tools to your account. The assistant shows every recipient, amount and total cost, and nothing is sent until you confirm.