Real Impact of Agentic Commerce on Payments and Sales Across Industries
The topic of agentic payments remains hot in the media and industry discussions, but some of the information currently circulating is already outdated. When we first analyzed it, ChatGPT was just launching its Instant Checkout, and Google was introducing its Agent Payments Protocol in an announcement format. Six months have passed since then — a significant period in the payments industry — and enough has happened to warrant returning to this topic.
Today, we will break down a compelling question: which of the announced features are actually working, and which have remained mere concepts or failed to live up to expectations?
Substantial investments are being poured into this sector, which is why many industries are closely monitoring the evolution of agentic commerce. Adapting to shifts in conventional processes is nothing new for the fintech sector, yet a multitude of myths have accumulated around AI-driven payments. Today, we will address them one by one.
What Agentic Payments Mean for Cardholders?
Let’s start with a brief look back and once again establish the difference between a standard chatbot and an AI agent capable of executing purchases. A typical chatbot can recommend an item, compare prices, or provide product details, but its role ends there — the consumer must still navigate to the merchant's website, enter card details, and confirm the order (which is what most users do today). Conversely, an agent does not just offer advice; it drives the transaction to completion, including the payment itself. The buyer simply states their goal, and agent finds the right option, places the order, and confirms delivery without switching tabs or inputting payment credentials. In practice, it becomes a virtual assistant with access to the client’s wallet, capable of executing tasks autonomously.
For payments industry, this primarily signals a shift in the checkout location. Previously, the entry point was the merchant's website or a marketplace, and the competition centered on driving traffic to a specific store. Now, the entry point is shifting to the dialogue window of ChatGPT, Gemini, or any other agent, and the competition is moving there: who can embed their payment mechanism directly into this conversation so that the merchant maintains customer control while the payment network retains the transaction? This is precisely why, over the past six months, literally all major market players — payment networks, AI developers, and large enterprise merchants who desperately want to avoid being left out of this new sales channel — have entered the race simultaneously.
To enable an agent to pay without giving it unrestricted access to a user’s full card balance, the industry has devised several conceptually similar mechanisms. They carry different names across companies but solve the exact same problem. Stripe, in its integration with OpenAI, utilizes a token restricted to a specific merchant and purchase amount; here, the agent does not receive the full card data, but rather a single-use key valid only for that specific transaction. Google structured this as a set of signed “mandates,” where the user's intent, the agent's specific product/price proposal, and the financial component seen by the payment network are all logged separately. The underlying logic is identical everywhere — the agent operates within strictly defined boundaries rather than gaining full discretion over the user's funds. Furthermore, across all major protocols, the merchant remains the entity that processes the order, calculates taxes, and handles fulfillment, while the agent acts as an intermediary positioned between the consumer and the business.
Which Agentic Payments Are Already Live?
While Google had only announced its Agent Payments Protocol back in November 2025, the landscape has shifted substantially since then. At its May conference, Google showcased Universal Cart — a unified shopping cart that operates seamlessly across Search, Gemini, YouTube, and Gmail, backed by a catalog of over sixty billion product listings. This functionality is already partially live in the US, working with selected brands (such as Nike, Sephora, Target, Walmart, and Wayfair, as well as Shopify-based stores). The architecture closely mirrors the token logic used by Stripe but is deployed within Google's own ecosystem. A user can pre-set a condition for the agent, such as “buy this exact model, provided it's under a certain price point.” Once the conditions are met, the agent independently generates the invoice and completes the purchase, while the entire decision tree is captured as a verifiable record that can be referenced in the event of a return or dispute.
Payment networks have also transitioned from roadmaps to live transactions. For instance, Mastercard expanded its agentic network to Hong Kong, where users can already book and pay for an airport ride via an AI agent, with HSBC serving as the partner bank; similar pilots have been conducted in Australia, the US, and India. In Europe, Mastercard has taken a different route, focusing on horizontal infrastructure. While a turnkey consumer use case is not yet live in the EU, the baseline infrastructure is fully ready: in March, Mastercard and Santander executed their first controlled pilot, and by June, all issuing banks in Europe were connected to Agent Pay at the network level. This establishes a wide footprint upon which products can be built without being tied to a single banking institution.
Visa released Intelligent Commerce Connect, a product that offers merchants and payment providers a single integration point for multiple agentic commerce protocols simultaneously. This includes Visa's own Trusted Agent Protocol, alongside protocols from Mastercard, OpenAI, and Google. This addresses a common bottleneck seen in other areas of the payments industry: businesses no longer need to maintain five distinct integrations for five different agents; one entry point is sufficient. In parallel, Visa announced a partnership with OpenAI, directly embedding Visa payments into ChatGPT with tokenized card data, real-time authorization, and user-defined spending limits.
What’s Not Working and Why
Despite the undeniable pace of technological advancement, practical execution hasn't gone entirely according to plan. The most telling example is OpenAI, which aimed for a rapid scaling of its Instant Checkout but ran into unexpected hurdles. The initial plan involved onboarding merchants on Etsy, followed by a rollout to a million Shopify merchants. In practice, only about a dozen real merchants made it to launch. While the underlying logic remained sound — users did indeed use ChatGPT to research product options — they ultimately chose to complete purchases elsewhere, preventing the feature from reaching infrastructure-level volume. OpenAI did not even manage to configure automated sales tax collection and remittance, given the lack of meaningful transaction volume through the feature. As a result, the company pivoted its approach, shifting the focus away from native in-chat checkouts toward deep-linking into the applications of major partners like Instacart, Target, and Expedia, while reserving the protocol for a select group of enterprise merchants rather than mass-onboarding long-tail retail shops.
This situation highlighted an industry-wide challenge that extends beyond OpenAI: integrating with any of these protocols demands significant operational effort from the merchant. They must maintain a real-time product catalog, configure fraud prevention specifically tailored for automated orders, and ensure seamless inventory synchronization. While major networks like Visa and Mastercard can afford to negotiate with dozens of enterprise partners simultaneously and supply ready-made infrastructure, millions of small and medium-sized merchants simply cannot keep up with this pace. Consequently, mass adoption has been pushed out significantly further than initially projected.
On the regulatory front, progress has also failed to match the velocity of the technology itself. The EU AI Act is gradually coming into force — with core requirements for high-risk systems set to take effect this August — but harmonized technical standards specifically governing agentic systems have been deferred until the end of the year and remain unfinalized. Experts working directly in this field openly state that the legislation was drafted without a clear understanding of how autonomous agents actually behave, leaving critical questions regarding human oversight, data protection, and liability distribution between the agent developer, payment provider, and merchant entirely unresolved.
In the US, a unified federal framework is still absent, and the change in administration has introduced further uncertainty by rescinding previous secure AI directives without establishing a clear replacement. In practice, implementing agentic payments right now means operating under high regulatory ambiguity, with the understanding that the rules could shift dramatically. Furthermore, the vast majority of merchants lack a clear onboarding pathway, as direct integration with networks remains fragmented, meaning businesses will likely have to wait until major payment processors begin offering this as a standard turnkey service.
Practical Takeaways for Payments Industry
Ultimately, payment networks and big tech companies are moving far faster than anticipated, yet genuine market demand for this service remains low. Mass, ubiquitous onboarding of small and medium-sized businesses to these protocols is not happening yet, as infrastructure and regulations lag behind the speed of press releases.
For any business considering an integration with one of these protocols, the most prudent approach is to assess which specific transactions are already running live in production with a given partner. It is also critical to bear in mind that, unlike baseline card transactions, new dispute and fraud frameworks for agentic commerce have yet to be codified. Chargebacks and customer disputes in traditional e-commerce already cause significant friction for merchants, despite the payment method being mature and highly optimized. With the advent of agentic payments, merchants will inevitably face bad-actor consumers looking to exploit the system by claiming, “I didn't buy this; it was an unauthorized bot action.” Consequently, heavy emphasis is being placed on agent activity logging and multi-step approval chains. Nonetheless, it will take years to bring this process to the same level of institutional maturity that governs modern digital commerce today.
Even so, agentic commerce is no longer a distant futurist concept or uncharted territory; it is an active, understandable operational process. While consumers do not yet trust the technology completely, this sentiment is poised to evolve rapidly.
If you want to stay ahead of these shifts as they unfold rather than catching up every six months, subscribe to our Telegram channel, where we analyze these developments in real time.
