Stripe’s John Collison says agentic commerce will completely transform online shopping

John Collison thinks keyword search is a “ridiculous” way to find things to buy. The Stripe co-founder told Bloomberg that agentic commerce, in which AI agents shop on behalf of consumers, will completely transform the online shopping experience, reshaping not just how people purchase but how retailers sell.

The argument is structural. For more than a decade, e-commerce has been built around targeted ads, algorithmic recommendations, search engine optimisation, and infinite scrolling, a system designed to capture human attention and convert it into transactions. Agentic commerce replaces the human in the loop. When an AI agent evaluates products, compares prices, checks reviews, and initiates a purchase on a consumer’s behalf, the entire advertising and discovery infrastructure built for human eyeballs becomes less relevant. Brands will need to appeal to AI agents as well as, or instead of, human buyers.

Collison’s perspective is informed by Stripe’s position at the centre of internet payments. The company processes transactions for millions of businesses and has been building infrastructure specifically designed for agent-to-agent commerce. At Stripe Sessions 2026, held in San Francisco last month, the company unveiled its Agentic Commerce Suite, live integrations with Meta, Google, OpenAI, and Microsoft, alongside a Machine Payments Protocol co-authored with its blockchain subsidiary Tempo that enables AI agents to pay each other in stablecoins or fiat currency. Amazon responded this week by putting its Alexa for Shopping agent inside the main Amazon.com search bar, a defensive move designed to keep the buy flow inside Amazon’s ecosystem before external agents capture the high-intent query.

The question Collison raised in the Bloomberg interview, whether AI agents can truly mimic human taste, cuts to the heart of agentic commerce’s limitations. For commodity purchases, groceries, toiletries, repeat orders, an agent optimising for price, speed, and past preferences is straightforwardly useful. For high-consideration purchases, fashion, furniture, electronics, the role of personal taste, aesthetic judgment, and the experience of browsing is harder to delegate. The technology is advancing rapidly, but the gap between an agent that can find the cheapest flight and one that understands why you prefer a window seat on the left side of the aircraft is not trivial.

China is already further along this trajectory than the West. Alibaba integrated its Qwen AI assistant with Taobao’s catalogue of more than four billion products, reaching 300 million monthly active users. Alipay processed 120 million AI-agent transactions in a single week in February. Meituan, JD.com, ByteDance, and Tencent are all deploying similar capabilities. The structural advantage of Chinese super-apps, which integrate discovery, communication, payment, and fulfilment within a single environment, means the entire agentic shopping workflow can happen without leaving the platform. In the West, the buy flow still typically crosses multiple apps and websites, creating friction that agents must navigate and that incumbents can exploit.

The implications for retailers are significant. If an AI agent is the primary buyer, search engine optimisation gives way to something closer to agent optimisation, the discipline of making products legible to AI systems rather than to human browsers. Product descriptions, structured data, pricing transparency, and return policies all become inputs that agents evaluate programmatically. A brand that ranks well on Google but poorly in a ChatGPT shopping query may find its traffic evaporating.

Stripe is positioning itself as the payment infrastructure for this transition. Its Link product, which now has 250 million consumer wallets, has been adapted to function as an agent wallet, allowing AI agents to spend money on a user’s behalf within boundaries the user sets. Google, Amazon, and OpenAI are all building their own agentic commerce protocols, and the competition to control the payment rail that agents use is intensifying. Stripe’s bet is that it can be the neutral infrastructure layer that all agents transact through, regardless of which AI company built them.

Collison has previously described agentic commerce and stablecoins as “twin revolutions in intelligence and money.” At Stripe Sessions, William Gaybrick, Stripe’s president of product, used the same framing. The company’s $159 billion valuation, confirmed in a recent tender offer, reflects investor confidence that Stripe can capture value from both transitions simultaneously. Whether that confidence is justified depends on whether agentic commerce reaches the scale its proponents predict, or whether it remains, for the near term, a compelling idea that works better in conference keynotes than in the messy reality of online shopping.

The enterprise software industry is already restructuring around the assumption that agents will handle an increasing share of commercial activity, from procurement to customer service to payments. Collison’s argument is that retail will follow the same path, and that the companies that adapt their products, their data, and their payment flows for AI buyers will outperform those that continue optimising for human ones. The timeline is uncertain. The direction, he believes, is not.