Stripe's John Collison on How Agentic Commerce Will Reshape the Internet

Summary of Stripe's John Collison on How Agentic Commerce Will Reshape the Internet

by Bloomberg

47m•May 16, 2026
Listen to Episode

Overview of Odd Lots: Stripe's John Collison on How Agentic Commerce Will Reshape the Internet

This episode of Odd Lots explores the rise of agentic commerce—the idea that AI agents will increasingly research, compare, and even complete purchases on behalf of users. Bloomberg’s Joe Weisenthal and Tracy Alloway speak with John Collison, co-founder and president of Stripe, about how this shift could change online shopping, website design, advertising, microtransactions, stablecoins, and the broader structure of the internet. The conversation centers on a major thesis: AI won’t eliminate commerce, but it may radically reduce friction and reorganize how products, services, and information are bought and sold online.

What “Agentic Commerce” Means

Collison defines agentic commerce broadly as AI buying something for you. He breaks it into two main categories:

Consumer-facing agentic commerce

  • A user researches something in ChatGPT, Claude, or another AI app.
  • The AI helps narrow options, compare trade-offs, and potentially complete the purchase.
  • In the near term, this is mostly about removing checkout friction rather than giving AI full autonomy.

B2B / developer-led agentic commerce

  • AI tools like Claude Code may need to buy resources automatically as part of doing work.
  • Example: an AI agent buying and setting up a domain name or paying for cloud services.
  • This is a big deal for developers because it removes annoying setup steps and form-filling.

Stripe’s View: The Biggest Change Is Friction Removal

Collison argues that the history of online payments has been a long march toward less friction:

  • credit cards
  • PayPal
  • Apple Pay / Google Pay
  • one-click and tokenized payments

Agentic commerce, in Stripe’s view, is the next step:

  • the agent fills out forms
  • the user still makes the final decision
  • the purchase becomes faster and easier

He emphasizes that this is not necessarily about AI “deciding” what to buy. More often, it’s about:

  • researching options
  • surfacing better products
  • letting the user approve the final purchase

How Shopping, Discovery, and Taste May Change

A major thread in the discussion is how AI changes product discovery.

Better search than keyword boxes

Collison is skeptical of traditional keyword search for many shopping tasks:

  • it works for exact items like books or DVDs
  • it is much worse for furniture, clothes, niche products, or anything with constraints

AI-driven shopping can handle:

  • dimensions
  • use case
  • budget
  • product trade-offs
  • contextual needs

Smaller brands may benefit

Because AI can surface products based on fit rather than brand recognition:

  • smaller merchants may become more discoverable
  • niche products can surface more easily
  • the market may become more dynamic

But humans still matter

Collison pushes back on the idea that AI will replace human taste entirely:

  • people still enjoy browsing, scrolling, and discovery
  • for many categories, the search process is part of the fun
  • humans will likely remain the final deciders in many purchases

What Businesses Will Need to Do to Serve Agents

The interview spends a lot of time on the practical side: how merchants make their products usable by AI agents.

Machine readability

Businesses need to make their offerings legible to agents by:

  • having clear product pages
  • publishing detailed specs
  • making reviews accessible
  • exposing live inventory and pricing information

Real-time product data

A model may know about a product from its training data, but it still needs:

  • current stock levels
  • current sizes/SKUs
  • current pricing and sale status

Agentic checkout infrastructure

Stripe is building tools so agents can:

  • securely access payment credentials
  • complete purchases without exposing sensitive card data
  • transact across systems with less manual user intervention

Bot protections need to evolve

Historically, websites defended against bots. Now, Collison says, businesses may need a form of:

  • “TSA PreCheck for bots”
  • trusted agent access
  • secure, controlled one-time-use credentials

Website Design: Built for Humans or Bots?

One of the biggest conceptual questions is whether the internet will become:

  • more human-friendly, with AI agents navigating existing sites
  • or more machine-native, with websites designed primarily for agents

Collison says both paths are possible:

  • AI may get good enough at unstructured computer use to operate any site
  • or the web may evolve to expose more textual, structured interfaces for agents

Stripe is currently betting on making its own systems more consumable by AI, especially through text-based interfaces and programmatic access.

Advertising, Brand, and the Future of the Internet

The conversation goes beyond payments and into the internet’s economic model.

Advertising is not going away

Collison is skeptical that agentic commerce will create a post-advertising world:

  • humans still make final decisions
  • brand preference still matters
  • undirected shopping will continue on social platforms and discovery feeds

But ad mechanics may change

Some implications:

  • less traffic flowing through traditional search
  • more relevance in AI-generated product recommendations
  • potentially less value for low-quality SEO content
  • strong brands may remain powerful, but the path to discovery changes

Culture and commerce remain linked

Joe and Tracy push the idea that advertising does more than sell products—it shapes culture. Collison agrees that:

  • brand identity still matters
  • ads will likely persist as long as humans remain involved in commerce

Microtransactions and Stablecoins

A major theme is whether AI will finally make tiny payments practical.

Why microtransactions have historically failed

Small payments have long been impractical because:

  • the user decision load is too high
  • card processing fees are too expensive
  • subscriptions and bundles are easier

AI changes the equation

If an AI can make hundreds of tiny decisions cheaply, microtransactions become more feasible:

  • pay-per-query for data
  • pay-per-use for APIs
  • pay-per-request for content or services

Stablecoins could help

Collison says Stripe is interested in stablecoins because they can:

  • reduce payment friction
  • support small, usage-based charges
  • let users pay across many services without creating separate billing relationships everywhere

He gives examples like:

  • paying for a single financial data request
  • paying per web request
  • paying for services used in AI workflows

Data Licensing and the Open Web

The episode also touches on how AI affects the economics of online content.

Free web vs. licensed data

Collison argues AI is not ending the free internet, because:

  • much of the internet has always been paywalled or proprietary
  • many valuable data sources already operate on licensing models

AI may increase demand for licensing

He expects more deals like:

  • content licensing for AI training and retrieval
  • data partnerships with publishers
  • paid access to specialized datasets

Human browsing still exists

He distinguishes between:

  • human access to sites like Reddit or X
  • AI training or scraping access, which may increasingly be monetized

Liability, Trust, and Human-in-the-Loop Controls

A recurring concern is: what happens when the agent gets it wrong?

Collison’s answer is that most useful agentic commerce will still keep a human in the loop:

  • the agent can recommend
  • the user approves
  • spending limits can be set
  • delegated authority can be controlled like in a company budget process

He suggests that liability concerns are often overstated because the model is not “fully autonomous” in the important cases Stripe is building for.

Stripe’s Internal Use of AI

The interview also covers how Stripe itself is using AI.

Engineering

  • AI is heavily adopted in software engineering
  • some bugs are fixed via AI tools without humans opening an editor
  • this is the strongest and most mature use case

Sales and go-to-market

  • sales is measurable, so AI productivity is easier to track
  • Stripe’s sellers are enthusiastically using AI tools

Legal, finance, and risk

These are more difficult because:

  • the data is harder to structure
  • the models are weaker with specific internal financial data
  • security and access controls matter more

Collison’s broader point: AI capability is uneven, and companies need a realistic sense of where models are strong and where they still fail.

Bigger Picture: More Dynamism, More Entrepreneurship

Collison’s optimistic long-term view is that AI and agentic commerce will lead to:

  • more startup formation
  • more small businesses
  • more competition
  • lower barriers to launching and operating companies

He cites Stripe data showing:

  • new business creation up 71% year over year in Q1
  • app launches rising after years of stagnation

His thesis is that AI will make it easier for individuals and smaller firms to:

  • build products
  • compete with incumbents
  • coordinate work
  • experiment quickly

Key Takeaways

  • Agentic commerce is mostly about reducing friction first, not eliminating human choice.
  • AI will likely reshape shopping by improving discovery, comparison, and checkout.
  • Businesses will need to become more machine-readable and expose real-time product data.
  • The internet may evolve toward a hybrid model: human browsing plus agent-native transactions.
  • Advertising is likely to persist, though its role and mechanics may change.
  • Microtransactions and stablecoins could finally make tiny payments practical.
  • Stripe sees agentic commerce as part of a broader shift toward more economic dynamism and more entrepreneurship.

Bottom Line

The episode frames agentic commerce as a foundational shift in how the internet works: not just a better checkout button, but a new layer where AI systems help users discover, evaluate, and buy things. Collison’s core belief is that the internet’s next era will be defined by lower-friction transactions, smarter product discovery, and more programmable commerce infrastructure—with humans still making the ultimate decisions, at least for now.