Earlier this year Google released shopping features as part of its new AI Mode, with Google enabling AI agents to spot the best deal for a specific product and complete the checkout for the user. This is just one example of the early rise of “Agentic Shopping” – AI agents that aren’t just recommending products but making purchases on the consumer’s behalf.

These agents can monitor inventory across retailers, track price changes, compare deals, and execute transactions when conditions are right. It’s the difference between asking your AI “where can I find this?” and your AI simply telling you “I’ve found it and bought it for you.”

Let’s say I’m looking for a specific hairdryer (I wish!). I tell my AI agent: “Let me know when the TQQ hairdryer drops below £40.”

Here’s what happens next:

• My agent monitors multiple retailers continuously for the hairdryer I’m interested in
• When the price drops to £40 at a specific retailer, I get an alert from my agent
• I approve the purchase with a simple confirmation
• My agent completes the checkout, using my saved payment preferences I receive an order confirmation without opening a browser*

*With Perplexity and OpenAI introducing their own browsers (Comet and Atlas respectively), these purchases can also be completed in-browser.

This is agentic shopping in action. But for this to work seamlessly across different AI assistants, retailers, and payment systems, we need shared protocols so that all agents in these workflows speak the same universal language.

The protocols making this possible

OpenAI recently introduced its Agentic Commerce Protocol (ACP), an open standard that enables conversations between buyers, their AI agents, and businesses to complete purchases. ACP creates a universal language that allows ChatGPT to have shopping conversations with merchants, handling things like order placement (“complete this purchase”), product discovery (“show me the options”), and availability checks (“is this in stock?”).

Google has introduced its own Agent Payments Protocol (AP2) to tackle the payment side of agentic shopping, focusing on three key safeguards: proving users have authorised specific purchases (authorisation), verifying that agent requests accurately reflect user intent (authenticity), and determining accountability when fraudulent or incorrect transactions occur. Whilst ACP handles the shopping conversation, protocols like AP2 and payment integrations like PayPal’s wallet integration handle the money movement and security.

PayPal: adding the payment layer

Earlier this week, PayPal announced a deal with OpenAI to include its digital wallet in ChatGPT. Users will be able to purchase through ChatGPT by clicking a “Buy with PayPal” button, and PayPal merchants can sell their products through ChatGPT.

Here’s the crucial bit: when ChatGPT completes a purchase on your behalf using ACP, it needs a way to actually pay the merchant. The PayPal wallet functions as a payments layer on top of ACP, ensuring transactions are completed securely on behalf of the user:

• Stored payment credentials – Your payment method is securely saved
• Transaction execution – The actual money movement happens through PayPal
• Authorisation proof – PayPal can verify you’ve authorised this specific agent to make purchases
• Fraud protection – PayPal’s existing security infrastructure protects the transaction

Here’s how the complete flow works:

1. I tell ChatGPT: “Buy me the TQQ hairdryer I looked at”
2. ChatGPT uses ACP to communicate with the merchant (check availability, get price, initiate order)
3. When it’s time to pay, ChatGPT uses my PayPal wallet to complete the purchase
4. The merchant receives confirmation of successful payment through ACP
5. I get my order confirmation

What this means for Product Managers

For any product manager building ecommerce features, agentic shopping introduces a new paradigm for how users interact with your product:

• A simpler conversion funnel – Most traditional ecommerce funnels assume users browse, compare products, add items to cart, and check out. Agentic shopping can collapse this entire funnel into a single interaction. What does this mean for your product metrics? How do you measure success when users never visit your site?

• Users trusting shopping agents – With users delegating purchase decisions and checkout to AI, we need to ensure our products are transparent about what agents can and cannot do, and build in proper controls for users.

• New user needs emerge – As agentic shopping gains wider adoption, users will no doubt want to set personalised purchase criteria, manage multiple agents, and review agent actions. These are new product surfaces that don’t exist in traditional ecommerce.

• Increased importance of data and privacy – Agents need access to user preferences, purchase history, and payment information. We need to ensure products securely handle data sharing with AI agents whilst maintaining user privacy. What data do you share and what stays private?

Main learning point: Agentic shopping will change how we discover and buy things online. Open protocols like ACP and AP2 are laying the groundwork for AI shopping agents that can work across different platforms and retailers.

Related links for further learning:

1. https://agenticshopping.substack.com/p/the-age-of-agentic-shopping-has-arrived
2. https://openai.com/index/buy-it-in-chatgpt/
3. https://newsroom.paypal-corp.com/2025-10-28-OpenAI-and-PayPal-Team-Up-to-Power-Instant-Checkout-and-Agentic-Commerce-in-ChatGPT
4. https://www.cnbc.com/2025/10/28/paypal-openai-chatgpt-payments-deal.html
5. https://shopify.engineering/commerce-payments-protocol
6. https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol
7. https://www.bcg.com/publications/2025/agentic-commerce-redefining-retail-how-to-respond