Key Takeaways

  • John Collison, co-founder of Stripe, outlines five escalating levels of "agentic commerce," where software agents gain increasing autonomy in making purchasing decisions, moving beyond simple automation to reasoning and independent action.
  • Current technology sits primarily at Level One (simple automation like Meta's in-app checkout) and Level Two (AI shopping assistants that can reason through complex requests, like ChatGPT or Wayfair).
  • Collison demonstrated a live example of Level Three and beyond: an AI agent autonomously bought access to specific data using micropayments via stablecoins through Stripe's Tempo CLI, highlighting real-world, machine-to-machine transactions already happening.
  • The demand for these capabilities is already growing rapidly, evidenced by 125,000 downloads of payment-related skills on Claw in just 12 weeks.
  • Founders should recognize that this isn't a future trend, but an immediate opportunity; if your product or platform can support machine-to-machine payments, Collison advises building for it now to capture early-mover advantage.

The Five Levels of Autonomous Commerce

Forget simple chatbots. John Collison, co-founder of Stripe, sees commerce rapidly evolving towards a future where software agents don't just help you shop, they are shopping. He breaks this down into five escalating levels of autonomy, moving from basic assistance to full machine-to-machine decision making.

Today, most consumer interactions are at Level One: what Collison calls "simple help." Think of Meta's one-click in-app checkout or a basic "buy now" button. It's automation, but it still requires direct human intent and input at every step. It streamlines a known process.

Level Two represents a significant leap. This is the shift Collison describes “from plain old keyword search that we've had for decades to a shopping assistant that can actually reason within constraints and find products accordingly.” Imagine asking ChatGPT or a sophisticated assistant like Wayfair to "find me a durable, pet-friendly couch under $1,000 that ships in two days." The agent understands the nuance, checks multiple parameters, and suggests options. It's smart, but you're still making the final call.

The real game-changer is Levels Three, Four, and Five, where agency truly kicks in. Here, AI agents move from assisting to autonomously executing purchasing decisions. Collison illustrated this with a stark example: an AI agent was given a query for specific data. “The agent analyzed my question,” Collison explained, “It looked for the relevant sources. It found a paid source, and now it's off buying and downloading that data autonomously.” This wasn't a human clicking "buy"; it was software negotiating and paying for information with micropayments using stablecoins via Stripe's Tempo CLI. This machine-to-machine transaction happened without human intervention, from decision to payment execution.

The immediate demand for this agentic capability isn't just theoretical. Collison pointed to platform data from Claw, revealing “the cumulative downloads of payment-related skills on Claw, 125,000 in 12 weeks.” This explosion of interest shows that developers are actively experimenting and building for these autonomous payment flows, creating a rapidly expanding ecosystem for agents that can pay.

Where This Breaks Down (Today)

While the potential is huge, widespread agentic commerce isn't without its current friction. The primary challenge remains trust and oversight. Humans are still hesitant to cede full financial control to an AI, especially for larger, more subjective purchases. Early adoption will likely concentrate in micro-transactions, data procurement, or highly standardized B2B supplies where parameters are clear and risk is low. Regulatory frameworks also lag behind, creating uncertainty for founders building in highly regulated industries. Furthermore, the underlying infrastructure, while evolving rapidly with tools like Stripe's Tempo, is still early-stage. Integrating complex legacy systems with new machine-to-machine protocols requires expertise and investment that not every startup has access to.

What to Do With This

Stop planning for the distant future; start building for autonomous payments this quarter. First, audit your current customer journeys and product offerings: where could a non-human agent plausibly make a decision and complete a purchase? Perhaps it's automated API access, data subscriptions, or even smart home device resupplies. Second, get hands-on with tools like Stripe's Tempo CLI (or similar emerging platforms) to understand how machine-to-machine payments actually work. Pilot a small, low-risk project where an AI agent—even a rudimentary one—can autonomously acquire a digital asset or service from your platform. The goal isn't perfection, but to learn by doing and position your product for the inevitable surge of agentic commerce.