Stripe co-founder John Collison used a Bloomberg podcast appearance on May 16, 2026 to argue that agent-led purchasing will compress the entire online buying funnel, declared keyword search "ridiculous," and disclosed that new business creation on the payments platform rose 71% year over year in the first quarter of 2026.

Speaking on Bloomberg's Odd Lots with hosts Joe Weisenthal and Tracy Alloway in an episode published yesterday, the Stripe president was asked to define the term that has dominated payments and search discourse for the last eight months. According to Collison, agentic commerce is, at its simplest, "AI buying something for you." He distinguished two categories: a consumer side, where shoppers research products in ChatGPT or Claude and either complete the purchase themselves or delegate it to the model, and a business-to-business side, where coding agents transact for services such as cloud infrastructure and domain registration.

The conversation arrived three weeks after Cloudflare and Stripe published a protocol on April 30, 2026 letting AI agents provision accounts, register domains, and deploy applications without manual intervention. Collison referenced that integration directly, framing it as a working example of the B2B leg of agentic commerce.

Friction reduction as base case

Weisenthal pressed Collison on whether agentic commerce represents simply another step in the long compression of online checkout friction, or something more fundamental. Collison answered "both," but emphasized the friction-reduction layer as the dependable near-term outcome.

According to Collison, the general direction of travel in payments has run from the introduction of credit cards through early e-commerce form-filling to mobile wallets like Apple Pay, Google Pay, and Stripe's own product Link. He argued the next step in that sequence is letting the agent complete the form rather than the human. "Do you really then want to go and be filling out all these web form fields and things like that?" he said. "Or do you want to just say, yeah, that sounds good, buy it for me in this size." His thesis: "In the history of technology, the lower friction option tends to win out."

Stripe's commercial bet on this thesis is documented. The company co-developed the Agentic Commerce Protocol with OpenAI when ChatGPT's Instant Checkout launched on September 29, 2025. It participated in Microsoft's Copilot Checkout on January 8, 2026 alongside PayPal and Shopify. And on April 24, 2026, Stripe joined the Universal Commerce Protocol Tech Council alongside Amazon, Meta, Microsoft, and Salesforce, formally entering the governance body of a competing open standard.

The keyword search critique

Asked whether retailers are already restructuring product listings for AI agents rather than human shoppers, Collison offered the bluntest position of the conversation.

"Keyword search is ridiculous," he said. "It's ridiculous that we got to the year 2026 relying on keyword search where that makes sense for buying a book or a DVD, where you know the title of the book or DVD that you're trying to buy. But that's about the limit of keyword search."

According to Collison, the keyword text box at the top of every retail website was a structural anachronism that AI-mediated research naturally replaces. He gave the example of looking for a piece of furniture with specific physical constraints - a defined slot in a room with a maximum width and height - and noted that conversational search handles such queries more effectively than a text input field. The same dynamic, he argued, surfaces small merchants that traditional aggregators have historically buried.

That framing aligns with analysis published on PPC Land at the launch of the Universal Commerce Protocol on January 11, 2026, which documented how Google co-developed UCP with Shopify, Etsy, Wayfair, Target, and Walmart specifically to address conversational product discovery in cases where traditional keyword input failed. Google's head of Ads and Commerce, Vidhya Srinivasan, made a similar point on the Frontier CMO podcast on February 13, 2026, describing 2026 as the year the trade-off between speed and certainty in shopping begins to dissolve.

On advertising's survival

Weisenthal raised the question of whether agentic commerce signals the end of advertising as the dominant economic engine of the consumer internet. Collison rejected the premise on two grounds.

First, according to Collison, agentic commerce does not remove the human from the buying loop. He used a hypothetical podcast microphone purchase to illustrate: a shopper might ask an AI to research three or four options, retain a brand preference based on previous knowledge or exposure, and only then delegate the actual execution. "Brand affinity really matters in that world," he said. "And as part of that, advertising will clearly stick around."

Second, Collison drew a distinction between directed and undirected commerce. Directed commerce, where the consumer knows what they want and uses search to find it, may indeed see ad inventory migrate from Google text ads to AI app results. Undirected commerce, on the other hand, occurs on platforms like Instagram, where users scroll without specific buying intent. According to Collison, those advertising surfaces are unlikely to be disintermediated by agents.

The macroeconomic data tracks with that assessment. Alphabet reported its Q1 2026 earnings on April 29, 2026, with Google Network advertising revenue falling 4% year over year to $6.97 billion - the sharpest quarterly decline the segment has recorded in recent reporting periods. That fall has been concentrated in third-party publishers, not in Google's owned surfaces. Earlier data from July 2025 had shown the Network segment already in 1% decline as AI features kept users within Google's interface. Small publishers have absorbed disproportionate damage, with referral traffic from traditional search engines down 60% over two years for outlets generating between 1,000 and 10,000 daily page views, according to Chartbeat data published in March 2026.

Collison's analysis suggests media properties oriented toward human consumption will continue to do well with the advertising revenue they have, while what he described as "really, crummy, you know, aggregator pages" filled with Taboola ads and pop-overs may face harder times. "I don't think anyone's going to really be too sad about that," he said.

Machine readability and merchant infrastructure

Alloway asked Collison to describe what optimizing for AI agents actually requires of merchants. According to Collison, two layers of work are involved.

The first is presence in the training corpus. Large language models have read the public web. Product pages, third-party reviews, and editorial coverage such as Wirecutter all contribute to whether a model "knows" about a given product. That visibility, he said, functions similarly to traditional SEO.

The second layer is what Collison called the "mechanical wiring up" - the live API connections that let an AI verify a product is still on sale, in stock, and available in the requested size. This is the layer where Stripe's commercial work concentrates. The company is collaborating with Google, Microsoft, Meta, and OpenAI to build the protocols that move a confirmed shopping decision into a confirmed transaction.

Collison also described the inversion of historical bot policy. Websites that once aggressively blocked automated traffic now need to authenticate agents and route them through secure payment paths. Cloudflare, Visa, and Mastercard formalized that authentication layer on October 24, 2025 through the Trusted Agent Protocol and Agent Pay, both built on Web Bot Auth using HTTP Message Signatures with public key cryptography.

Which side wins: agent-readable websites or computer-use AI?

Weisenthal posed what Collison described as one of the active frontier debates in AI development: will the world build agent-friendly interfaces, or will AI models simply become good enough at navigating standard human interfaces using a virtual mouse and keyboard?

Collison disclosed where Stripe is allocating engineering resources. "Right now, a lot of people, including us, are working on making, you know, the Stripe dashboard, merchant website, things like that consumable by the AI," he said. Practically, that means converting interfaces previously built around clicking into text-based representations that language models can parse.

He acknowledged this is a wager. If models become proficient at unstructured computer use through a headless browser within the next two to three years, the investment in agent-specific interfaces becomes redundant. According to Collison, the case for doing the work anyway is that businesses moving first will capture demand before rivals: an AI app presenting two options, one with a Buy Now button and one requiring website navigation, will see consumers consistently choose the former.

Microtransactions and the stablecoin question

Alloway raised the perennial question of microtransactions - selling individual articles or data calls at fractional prices - and asked whether the AI usage-billing model finally makes that economically feasible.

According to Collison, the historical barrier to micropayments was not pricing but cognitive load. Selling Bloomberg articles for $0.40 each never worked because human decision-making about whether to buy was disproportionate to the cost. AI agents change that calculus. "Microtransactions are probably going to be now possible," he said, citing the example of pulling specific Bloomberg data for a financial query and paying a small fee per request.

The infrastructure for that pattern exists. Cloudflare launched its pay-per-crawl service on July 1, 2025, processing more than one billion HTTP 402 response codes daily, and built the system specifically to anchor a future where intelligent agents programmatically negotiate access to digital resources.

Stripe's interest in stablecoins, Collison disclosed, is rooted in the same architecture problem. He acknowledged real-world stablecoin usage in B2B agentic transactions remains limited. The opportunity, according to him, is in eliminating the need for users to maintain billing relationships with every service an agent might touch. Pay-per-request stablecoin settlement could remove that overhead entirely.

The usage-based billing model itself is now widespread in the AI economy. Collison noted that no provider offers an all-you-can-eat plan because marginal token costs prevent it. Stripe's 2024 acquisition of Metronome was structured specifically to power this kind of credit-based, telco-style consumption billing.

Running a payments company through compressed cycles

Weisenthal asked how Stripe allocates resources between exploiting current opportunities and exploring future ones when individual product categories can become obsolete in three to six months.

According to Collison, the company has settled on flexibility plus what he called calibrated belief in model capabilities. He singled out spreadsheet work as an illustration of where the limits remain stark. The models "can put together something that, you know, they can build a really nice spreadsheet that you're like, but the numbers just like go completely wrong halfway through."

He broke the company into three operational domains. Software engineering, where adoption has been fastest and where bugs at Stripe are frequently fixed without anyone opening an editor; sales and go-to-market, where measurability has driven aggressive adoption with productivity gains visible in numbers; and general and administrative functions - legal, finance, risk - where infrastructure work to expose internal data to AI systems has been substantial and results have been uneven.

The intelligence explosion thesis

Collison closed with a numerical claim that, if it holds, has implications well beyond Stripe.

According to Collison, new business creation on Stripe in the first quarter of 2026 was up 71% year over year. He compared the figure to App Store launch growth, which had been flat for years and is now showing similar acceleration. His read: AI is materially lowering the cost of starting a company.

"Our prediction is that you get more economic dynamism," he said, "which we're already seeing in our leading edge numbers." Smaller firms, more of them, coordinating with each other through agent-mediated commerce protocols. The reasoning loop closes: if the cost of building and launching is falling, and the cost of acquiring customers through agent-mediated channels is changing, then the structural shape of online commerce changes with it.

That conclusion sits in tension with the position of platforms whose commercial models depend on owning the customer relationship. Amazon has filed federal litigation against Perplexity AI over alleged unauthorized agent deployment, while Walmart, the original launch partner for OpenAI's Instant Checkout, has publicly described that product as having underperformed expectations. Google has begun extending UCP-based checkout from AI Mode into mainstream search results, with Wayfair listings the first documented case on May 5, 2026.

For now, Collison's framing is the most public articulation from a payments infrastructure principal of why the transition will happen regardless of whether individual retailers welcome it. According to him, the friction differential alone will determine the outcome.

Timeline

Summary

Who: John Collison, president and co-founder of Stripe, interviewed by Bloomberg's Joe Weisenthal and Tracy Alloway on the Odd Lots podcast. Stripe is the payments infrastructure company that co-developed the Agentic Commerce Protocol with OpenAI in September 2025 and joined the Universal Commerce Protocol Tech Council on April 24, 2026.

What: A wide-ranging conversation in which Collison defined agentic commerce as "AI buying something for you," called keyword search "ridiculous," predicted that brand advertising will survive the transition because humans remain in the buying loop, identified microtransactions and stablecoins as previously unviable models that the AI economy enables, and disclosed that new business creation on Stripe rose 71% year over year in the first quarter of 2026.

When: The episode was published on May 16, 2026, three weeks after Stripe and Cloudflare announced their joint agent provisioning protocol on April 30, 2026, and three weeks after Stripe joined the UCP Tech Council on April 24, 2026.

Where: The Odd Lots podcast, produced by Bloomberg, distributed through YouTube, Spotify, Apple Podcasts, and Bloomberg's own platforms. The episode was recorded after both Weisenthal and Alloway had attended an event in Madrid where the Stripe-Cloudflare announcement was made.

Why: For the marketing community, the interview matters because it offers the most detailed public articulation to date from a payments infrastructure principal of why agentic commerce will continue to expand regardless of resistance from individual retailers. Google Network advertising revenue is in decline. Small publishers have lost 60% of search referrals over two years. The advertising surfaces most exposed to AI mediation are exactly those Collison identified as "directed commerce" - keyword-driven shopping queries where consumers know what they want. The surfaces he identified as resilient - undirected social-feed advertising and brand-driven preference formation - align with the channels Forrester analysis cited by PPC Land predicts will absorb display budget cut from the open web in 2026. The 71% Stripe figure, if it holds, suggests the supply side of agent-mediated commerce is expanding at a pace that will force advertising and platform infrastructure decisions across the industry over the coming quarters.

Share this article
The link has been copied!