AI Shopping Agents Force Payments Firms to Rebuild Infrastructure

Delegation is replacing discovery as the defining characteristic of next-generation commerce. New research from PYMNTS Intelligence reveals that 48% of consumers are at least somewhat interested in artificial intelligence agents handling grocery shopping or meal planning, while an identical share would permit autonomous assistants to manage their subscriptions.

The shift represents more than incremental convenience. When artificial intelligence systems acquire spending authority, the center of gravity in commerce moves away from the checkout page toward infrastructure that governs how machines spend money on human behalf.

Traditional eCommerce models rely on influencing human intent: surfacing products, optimizing conversion funnels, and reducing cart abandonment. In an agentic landscape, these metrics lose meaning when carts are no longer assembled by humans. The battleground is no longer the moment of purchase. It is the authorization layer where rules are enforced and trust is negotiated between human, machine, and merchant.

Card networks, issuing banks, and FinTech platforms are rapidly repositioning themselves to capture this moment. Their goal is not merely to process transactions, but to define the frameworks through which artificial intelligence can transact safely and effectively. This involves more than enabling API-driven payments. It requires building systems that can interpret and enforce user intent at scale (a problem that has plagued users for years, frankly).

For example, an AI agent might be authorized to reorder household essentials automatically, but not to make discretionary purchases above a certain threshold. It might prioritize vendors with specific sustainability credentials or avoid subscriptions that do not meet predefined criteria. Embedding these rules into the payments infrastructure transforms it into a governance layer. The payment credential becomes a programmable instrument, capable of encoding policy, context, and constraints.

The physical reality of this shift is stark. For routine transactions, checkout may ultimately become a back-end process, invisible to the user and optimized for machine interaction. No more clicking through payment screens. No more typing card numbers. Just silent, asynchronous authorization flowing through APIs that evaluate risk, enforce spending limits, and complete transactions in milliseconds.

FIS executive Mladen Vladic, head of product for payment networks, told PYMNTS that agent-mediated purchasing could account for as much as $1 trillion in retail revenue in the United States by 2030. That is just four short years away.

When Walmart announced a partnership enabling purchases through Google's Gemini app, it signaled more than experimentation. It suggested that the world's largest retailer views agent-mediated checkout as inevitable. Similar partnerships, such as Target's collaboration with OpenAI, have underscored how quickly the agentic model is being legitimized.

According to PYMNTS Intelligence research, the payments layer emerges as the critical control point as AI agents take control of purchasing decisions. This is where authorization happens, where rules are enforced, and where trust is negotiated.

A separate report from PYMNTS Intelligence and FIS finds that issuer false declines contribute to roughly $30 billion in annual lost sales globally. That friction becomes even more problematic in agentic commerce environments, where artificial intelligence agents may eventually initiate purchases without the traditional cues associated with card-present transactions.

The data requirements for that future are extensive. Machine learning systems rely on transaction histories, risk indicators, spending patterns, behavioral insights, and contextual data to make effective decisions. The report found that 47% of organizations still struggle with poor-quality data that limits AI effectiveness in decision-making.

Behavioral data becomes one of the most important because artificial intelligence systems need to distinguish legitimate autonomous activity from fraud or anomalous behavior. Spending histories and transaction patterns help create models capable of recognizing when an AI agent is operating within expected consumer preferences. Credential and authentication data also become essential. In agentic commerce, AI systems may need to validate payment credentials, apply spending controls, and execute transactions without direct user intervention.

Against that backdrop, Vladic pointed to another initiative as a bridge between today's payments reality and tomorrow's agent-driven future: Smart Baskets. Developed by FIS, the concept aims to unify network-level assets to optimize transactions in real time, factoring in least-cost routing, SKU-level intelligence, and loyalty network pivots at the basket level.

Asked what agentic success for FIS would look like by the end of 2026, Vladic returned to the scale of the opportunity. "If we deliver 10% of the projected agentic volume through our agent eCommerce or Smart Basket channels by the end of this year, we're on the right path," he said.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The competition is not just about processing volume. It is about owning the interface between human intent and machine execution. After all, when AI gets a wallet, commerce does not simply become faster or more convenient. It becomes fundamentally restructured. The question is no longer how to win at checkout. It is how to become the system that checks out for everyone else.

Whether users actually pay for this convenience remains the real question. Most consumers will never see the infrastructure enabling their autonomous purchases. They will only notice when something goes wrong. And in a world where machines spend money on your behalf, errors compound faster than humans can fix them.

The payments industry is building for a future where checkout disappears. Whether that future arrives depends less on technology and more on whether consumers trust machines with their wallets. Time will tell if that trust is warranted.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn