AI Agents Are Now Shopping for You — and That Changes Everything
On Friday, OpenAI made the agents in ChatGPT a little less chatty and a little more useful. The company launched ChatGPT Wallet, a feature that lets its “dots” — the AI agents inside the app — browse products, help you decide what to buy, and complete the purchase on your behalf.
This is the moment agents stop being assistants that reply and start being workers that act. The rest of the story is not about better models or faster chips. It is about whether a merchant’s checkout page, a risk engine, and a payment network are prepared to accept a purchase that no human actually typed in.
What the Wallet actually does
Eligible users add a U.S.-issued Visa, Mastercard, or American Express card to the wallet, then ask their dot to find something and buy it. A one-time setup: you enter the card once, and the dot handles the rest on that card.
Before anything is charged, you review the items, the total, and the payment card, then approve with a security check — Face ID, a fingerprint, or a similar biometric. The wallet then charges the card through a payment token: a substitute credential with its own card number, expiry date, and security code, scoped to that single transaction. Neither the dot nor the merchant ever sees your real card details.
The details matter. The merchant receives what it needs to fulfil the order — the shipping address and contact information — while the user keeps their card’s existing rewards and protections, including the ability to dispute a charge. OpenAI says it will not take a share of sales at launch, which is a deliberate choice: it lets merchants see the wallet as a checkout option rather than a competitor taking a cut.
Agentic commerce just got a funding tailwind
The same day as the Wallet announcement, the agent infrastructure startup LangChain raised $300 million at a $3 billion valuation. That round, led by Alphabet’s CapitalG, more than doubles LangChain’s valuation from its October 2025 Series B ($1.25 billion) and brings total funding to $460 million. LangSmith, its agent-management platform, is already used by more than 7,000 organisations, including Nvidia, Toyota, and Salesforce.
The two announcements point at the same conclusion: the industry has stopped asking whether AI agents will operate in the real world, and started asking how they will be governed, paid for, and trusted when they do. LangChain is building the tooling that lets every company own its own agents. OpenAI is building the wallet that lets those agents actually spend. For the AI economics behind it all, see Anthropic’s $1.4 trillion IPO.
The real bottleneck is trust
Visa’s CEO put it on a recent earnings call with unusual plainness: “In agentic, trust is the biggest barrier to scale.” The quote is the whole story. A purchase is a high-stakes handshake between a stranger, a merchant, and a machine. The wallet addresses the obvious risks — real card details never leave the user’s approved flow, every purchase gets a biometric seal, and every transaction leaves an auditable record.
But risk systems are built for human behaviour. They flag agent-driven traffic as fraud. Merchants’ checkout flows are full of bot-detection rules. Some sites actively block purchases made by agents. The trust question was never academic: United Airlines banned AI agents from booking flights earlier this year. OpenAI has responded by publishing a Commerce Partner Playbook — tools and guidance for retailers that want to accept agent shoppers — and by naming Riskified as a partner to help merchants recognise legitimate agent purchases instead of declining or blocking them.
That is a telling detail: the bottleneck is not the agent, it is the merchant. The agent can buy. The question is whether the checkout will say yes.
What comes next, and what to watch
The initial rollout is deliberately narrow: U.S. ChatGPT Pro subscribers, 18 and over, with one eligible card at a time. That is the right order of operations — a controlled experiment before anything broader. The genuinely interesting part is the roadmap. OpenAI plans to let users set spending limits, restrict purchases by category, and pre-approve purchases within time windows. That is the bridge from “agent buys what you ask it to” to “agent handles the recurring stuff without checking in.”
A few things to watch in the weeks ahead:
- Merchant adaptation. The Commerce Partner Playbook only helps if retailers actually change their checkout flows instead of blocking agent traffic. Watch for a wave of “buy with ChatGPT” announcements.
- Pre-approval behaviour. Spending limits and category filters will be tested in the real world — a $50 per-category cap reads very differently on a $9 streaming subscription than on a $2,000 gadget.
- Fraud at scale. The payment token model survived its first purchase. The real test will be whether a pattern of agent purchases eventually trips a risk engine, and whether disputes can be resolved as cleanly as they are for human cardholders.
The Philippine angle
For Filipino households still settling into smartphone life, agentic commerce is not an immediate concern. But it is an infrastructure story, and the Philippines is good at building infrastructure. The payments rails, the risk engines, the merchant checkouts — these are the plumbing that every new layer of digital commerce eventually runs through, from the energy grid to the chip factories. AI data centers and the energy grid are where it all starts. If agentic commerce becomes mainstream, the Philippines will either build its own agents and platforms, or import the ones that already exist. The ground stop for AI in 2026 came in the form of a ground stop for flights. The lesson is the same: trust is the bottleneck, and whoever solves it first owns the next layer.
ChatGPT Wallet is not a gimmick. It is the first time a major platform has handed an AI agent a payment credential and asked it to act in the real economy. The technology is ready. The commerce world is still catching up. That gap — between an agent that can buy and a merchant that will accept the buy — is where the next wave of AI will be written.