ChatGPT is shifting from a tool that helps compare products to something closer to a personal shopper, following a new integration with payments giant Visa.
The idea behind the collaboration is straightforward: instead of users browsing, filtering, and checking out themselves, artificial intelligence agents could increasingly handle the heavy lifting—whether that’s buying sneakers, event tickets, or other everyday services.
Once a user links a Visa cards to ChatGPT, they can prompt AI agents and include constraints like budget, timing, and delivery preferences. The agent then surfaces options and, where appropriate, completes the purchase on the user’s behalf.
This is not the first time either company has ventured into agentic commerce. OpenAI and Visa have each been building toward more capable transactional AI systems, but the partnership could speed up how quickly agent-led purchase moves from experimentation into mainstream use.
Global Agentic Momentum
There is no doubt that global momentum behind agentic commerce is building. ING recently completed Europe’s first end-to-end agentic payment transaction on Mastercard’s network, demonstrating that existing payment rails can support AI-mediated purchases. Mastercard has also run similar pilots in India.
By and large, many leading tech and payments players are steadily building the underlying infrastructure to facilitate agentic commerce. For example, Amazon Web Services has introduced features such as Bedrock AgentCore Payments, designed to let AI agents pay for digital services like web content and APIs.
A Wave of Experimentation
Beyond a few flagship pilots, there’s been a wider push to standardize how AI agents initiate and complete transactions. Much of this activity was largely sparked by Visa and Mastercard’s launch of agentic commerce platforms last year, prompting a wave of startups and enterprise integrations.
But despite the pace of development, adoption hasn’t been frictionless. OpenAI previously experimented with Instant Checkout as an agent-led shopping solution, but lack of consumer interest and the potential for agent error led the company to shutter the service.
That early setback highlights the core tension in this space. There’s enthusiasm from infrastructure providers, which is strong, but it’s unclear how often users actually want AI to buy things for them, and whether these systems can consistently meet expectations around accuracy, control, and trust.