Agentic commerce 2025: how the landscape is evolving

Oana Ifrim, Lead Editor at The Paypers, offers her perspective on agentic commerce – now a central theme in nearly every industry conversation.

A year ago, I dismissed "agentic commerce" as yet another overused buzzword. The hype felt disproportionate to the reality on the ground, and it was difficult to see any real commercial substance behind it. But 2025 has been a turning point. In conversations with retailers, banks, platforms, and payment networks, a clear consensus has emerged: autonomous shopping agents are no longer a future concept; they’re rapidly becoming the default interface for digital commerce.

What has shifted isn’t only the maturity of the technology; it’s the behaviour of consumers. For decades, digital shopping meant browsing pages, comparing options, and making decisions manually. Today, agentic commerce is transforming that model. Intelligent agents now handle much of the heavy lifting: searching, evaluating, negotiating, and even completing purchases on behalf of the customer.

The impact is profound. For consumers, it introduces a more seamless and tailored buying experience. For businesses, it forces a rethinking of engagement models, distribution strategies, and the infrastructure required to support this new mode of decision-making. And as these agents take on more responsibility, they bring forward critical questions around trust, transparency, and user control – because empowerment only works when people believe the decisions made on their behalf genuinely reflect their preferences and intent.

The shift toward agent-driven commerce is unfolding faster than most expected. According to Adobe, over half of consumers expect to use AI assistants for shopping by the end of 2025, and the shoppers they bring behave very differently: AI-referred customers show 10% higher engagement and move through the purchase journey with far stronger intent. That momentum is accelerating: traffic to US retail sites from generative-AI browsers and chat interfaces grew 4,700% year-over-year in July 2025, with these users spending 32% more time on-site, viewing 10% more pages, and bouncing 27% less.

This isn’t a temporary spike. It’s pointing toward a structural shift in how commerce operates. Worldpay forecasts that autonomous AI shopping agents will manage around 9% of all US online purchases by 2030, representing USD 261 billion in spend.

And the trend is not limited to the US. In the UK, 40% of consumers say they would delegate routine, low-engagement purchases – like bills and household essentials – to AI agents, according to Checkout.com.

Worldpay’s UK outlook reinforces this trajectory, projecting that AI-powered shopping assistants could drive GBP 29 billion in ecommerce spend by 2030.

These shifts didn’t happen in a vacuum. Payments players lit the fuse. Visa, Mastercard, and PayPal each made decisive moves this year, not with marketing soundbites but with actual infrastructure. Mastercard has launched Agent Pay, a new payments program that enables AI agents to make secure, tokenised transactions on behalf of users, built on a trust framework that verifies and registers agents. Visa launched Visa Intelligent Commerce, an AI commerce platform to enable secure transactions in an AI-driven retail environment. PayPal released its Agentic Toolkit to let developers build autonomous shopping flows.

Mastercard has launched Agentic Pay for developers, partnering with Stripe, Google, and Antom to enable secure agentic transactions worldwide. The program will be available to all US. Mastercard cardholders by the holiday season, with a global rollout soon after.

In early October, Mastercard announced that, alongside payment platform PayOS, it had completed a transaction using an agentic token that enables AI agents to execute payments on behalf of users: 

Two weeks later, Visa unveiled its Trusted Agent Protocol – a foundational framework for agentic commerce designed to establish secure communication between merchants and AI agents during transactions.

OpenAI has also moved decisively into this space with the launch of Instant Checkout, a feature allowing users to buy products directly within ChatGPT. Initially rolled out to US-based Etsy sellers and Shopify merchants in September, the service expanded last month through partnerships with Walmart and PayPal.

Mastercard has partnered with PayPal to contribute to the adoption of secure global agentic commerce.

How platforms are responding

A growing number of retail and commerce platforms are moving beyond traditional recommendation engines and into true agentic commerce – AI systems that can take action, complete tasks, and even execute purchases within defined guardrails.

Amazon is testing a "Buy for Me" feature in its Shopping app that uses agentic AI to let customers discover and automatically purchase select items from other brand websites (not sold on Amazon), handling checkout, order tracking, and customer data securely.  Most recently, Amazon is rolling out Rufus, a next-generation AI shopping assistant built on agentic and generative AI (via Amazon Bedrock), which learns from your past behaviour to offer personalised recommendations, answer detailed product questions, and even act on your behalf. Rufus can track price history, set alerts, and auto-buy when a target price is hit — using your default payment and shipping info.

Walmart has launched Sparky, a generative AI assistant designed to evolve into a fully autonomous shopping agent. Today, it supports search and recommendations, but Walmart’s roadmap includes “super agents” that can eventually plan and execute complex shopping tasks with minimal user input.

Adobe is expanding its enterprise AI capabilities with a new generation of agents designed to automate customer-experience, marketing, and merchandising workflows. Built on the Adobe Experience Platform (AEP), the system connects real-time customer data, content, and processes across the Experience Cloud. At its core is the AEP Agent Orchestrator, a reasoning engine that enables enterprises to build, manage, and coordinate specialised AI-driven agents that interpret user intent and execute multi-step tasks using both Adobe and third-party data and models.

Salesforce Commerce Cloud is pushing deeper into agentic workflows through Agentforce (Buyer Agents, Merchant Agents, and Personal Shopper Agents) that automate tasks like B2B reordering, customer service, and guided shopping with enterprise-grade guardrails.

Perplexity has integrated PayPal to enable native checkout within its chat interface. Users can browse, select, and purchase products directly in Perplexity, with PayPal managing payment, shipping, tracking, and confirmations. The integration leverages PayPal’s secure wallet and passkey-enabled checkout, streamlining the entire transaction process.

The key phases of agentic commerce

According to BCG, agentic commerce unfolds in four distinct phases, each building on the last. Here are some examples I’m seeing for each phase.

The first is product discovery, where agents help customers find and compare products but still hand the transaction off to the retailer’s website. This is already mainstream. ChatGPT can pull recommendations directly from Shopify stores. Google’s Shopping Graph surfaces and compares products across huge retail inventories. Amazon’s Rufus does the same inside the Amazon app, guiding customers through conversational prompts before sending them to the product page. In this stage, the agent shapes intent, but the checkout remains traditional.

Next comes purchase-in-chat, where the entire transaction happens inside the conversational interface. WhatsApp already supports this in India and Brazil, allowing users to browse and pay without ever visiting a website. Klarna’s ChatGPT experiences do the same, pulling users straight from product recommendations to in-chat checkout. TikTok Shop has also normalised this behaviour, letting people discover and buy in one uninterrupted flow. Here, the website becomes optional — the agent manages discovery and checkout in one place.

Right now, we are mostly in the guided discovery and early conversational checkout phases.

Moving further, autonomous purchasing sees agents handling almost the entire buying process independently. Alexa’s automatic reorders are a familiar version of this: users set preferences once, and the system restocks essentials when supplies run low. Instacart’s newer AI tools can assemble full grocery baskets with minimal human input. Developers experimenting with AutoGPT-style commerce agents are beginning to show how an AI could independently search, compare, select, and purchase based on budget and constraints. In this model, users shift from “deciding” to simply approving.

The most transformative evolution is agent-to-agent commerce, where buyer and seller agents autonomously negotiate, transact, and confirm payments without any human involvement. Early research at OpenAI explores how autonomous agents could negotiate directly with each other. Stripe and PayPal have both experimented with frameworks where agents execute verified transactions under programmable constraints. In enterprise environments, autonomous supply chain pilots already show agents sourcing inventory, assessing suppliers, and completing orders on behalf of organisations.

Across all stages, the pattern is clear: the locus of commerce is moving away from websites and toward intelligent systems acting on behalf of users and businesses. We’re not fully in the autonomous era yet, but we’re much closer than most of the industry realises.

While these four stages provide a solid framework, different industries and markets show more varied and nuanced patterns of adoption.

Hybrid human-agent collaboration is common in sectors where buying decisions are complex or high-stakes, such as luxury goods, financial services, or real estate. Here, AI agents assist by preparing options, analysing data, and flagging opportunities, but humans remain deeply involved in the final decisions. For example, robo-advisors in wealth management suggest investment moves but let humans retain control over trades.

Subscription and replenishment automation is a crucial subcategory in consumer packaged goods and healthcare supplies. Agents here not only reorder based on usage patterns but also handle negotiations and contract management over time. Think of Amazon Dash’s integration with Alexa or specialised pharma supply chains that use AI to maintain clinic inventory automatically.

Social and community-driven agentic commerce is thriving in fashion, gaming, and entertainment markets. AI agents factor in social proof, influencer trends, and community signals to recommend and sometimes automatically acquire trending or limited-edition products. Platforms like TikTok Shop merge AI with influencer content, making agent recommendations feel culturally relevant and timely.

B2B procurement and supply chain automation is where agentic commerce is rapidly gaining ground in industrial sectors. Here, autonomous buyer and supplier agents negotiate complex contracts, optimize sourcing, and manage supply risks, all within programmable guardrails. Enterprise platforms like SAP Ariba and pilot projects with Stripe or PayPal showcase this trend toward agent-led procurement.

Regulated financial services automation adds another layer, with AI agents operating under stringent compliance and transparency rules. These agents may autonomously execute trades, purchase insurance, or manage payments, but embed regulatory checks and audit trails to satisfy legal requirements.

These stages aren’t rigid or linear; rather, they overlap and evolve differently depending on market dynamics, regulatory environments, and consumer preferences. Retail and CPG markets often race toward autonomous purchasing, while financial services emphasise collaboration and compliance. Social commerce markets blend agentic automation with human influence, and B2B sectors push the frontier of agent-to-agent commerce faster than many expect.

The legal and technical challenges of agentic commerce

Where all of this gets complicated (and far more interesting) is in the unresolved questions. Agentic commerce breaks the traditional model of liability. If an AI agent buys the wrong thing, overspends, or misunderstands an instruction, who is responsible? The user? The platform? The merchant? The agent developer? Payment networks are beginning to define the rules, but regulators are not yet aligned, and the industry still lacks a consistent approach to responsibility, intent, and disputes for machine-initiated purchases.

Risk and fraud models also need to evolve. The signals we’ve relied on for twenty years (device fingerprints, behavioural biometrics, click patterns) disappear when the "customer" is a software agent. Payments teams have to rethink authentication, agent identity, delegated permissions, and spending boundaries. Agent impersonation will likely become a new category of fraud.

The infrastructure challenges are equally significant. Agentic commerce depends on secure identity frameworks, tokenised credentials, real-time decisioning, and new authentication standards like passkeys and delegated verification. This is a major reason why the networks are investing so heavily: the future customer isn’t a person navigating a browser. It’s a computing process making decisions on behalf of a person.

Regulators will shape adoption as much as technology does. The EU’s AI Act introduces tough obligations for systems that make financial decisions. The UK is beginning to explore consumer protection issues around autonomous transactions. The US CFPB is already analysing AI’s impact on financial decision-making. Agentic commerce will require new forms of auditability, explainability, and user consent that go far beyond today’s T&Cs.

The commercial implications are just as dramatic. Platforms that own the agent interface will control discovery, intent, and even brand preference. Payment networks will strengthen their position as the trusted layer underneath autonomous transactions. Wallets – PayPal, Amazon, Apple – are well-positioned to become the default payment containers for agents. Meanwhile, retailers that depend on on-site UX risk losing visibility as agents bypass websites entirely. Search and ad ecosystems will feel that impact too.

As we look to 2026, the trajectory is clear. Retailers will start launching their own branded shopping agents. Networks will introduce dispute rules designed specifically for AI-initiated transactions. Agent-to-agent B2B purchasing will move from pilot projects into practical implementations. And a meaningful share of ecommerce journeys will no longer involve a checkout page at all.

Agentic commerce isn’t an add-on to ecommerce. It’s becoming the operating system beneath it—one in which human decision-making is optional, and the buyer is often not a person but a process. The shift is already underway. The infrastructure is forming. And the organisations that understand this early will be the ones shaping the next decade of digital commerce.

I’ll be back soon with deeper insights on what agentic commerce means for banks and retailers, plus some insights on some of the most pressing questions. Stay tuned!

In the meantime, you can check out this conversation with payments expert Andrew Dresner, where we discussed how agentic commerce could reshape traditional payment flows, card acceptance economics, and several other key areas across the ecosystem.

About Oana Ifrim

Oana Ifrim is Lead Editor at The Paypers, with a strong passion for content planning, strategy, and industry research. She manages features covering fintech, banking, and payments modernisation, while ensuring accuracy and clarity throughout content and editorial coverage. Oana conducts expert interviews and thought leadership content, moderates webinars and conference panels, leads research projects, reports, and whitepapers, and represents The Paypers at major industry events.

She can be reached at oana@thepaypers.com or on LinkedIn.