Agentic Commerce Is Not Autonomous Yet — But the Infrastructure Is Already Being Built

Agentic Commerce Is Not Autonomous Yet — But the Infrastructure Is Already Being Built By Jason T Wade Agentic commerce is one of those markets where the language has moved faster than the evidence. Depending on which announcement, vendor deck, or analyst report you read, AI agents are already becoming autonomous shoppers, payment networks are preparing for a machine economy, and software is about to replace humans as the primary commercial actor. That framing is premature. The more defensible conclusion, as of August 2026, is narrower and more consequential: Agentic commerce has reached the transaction-infrastructure stage. It has not yet reached broad autonomous commerce. AI systems can already discover products, compare alternatives, recommend merchants, construct carts, and in selected environments execute approved transactions. Stripe, Visa, Mastercard, Google, OpenAI, Coinbase, Cloudflare, Shopify, and others are building increasingly sophisticated payment, identity, authorization, and commerce layers around those systems. But there is a critical difference between an agent being able to execute a transaction and an agent being trusted with standing economic authority. That gap defines the market. I use a six-stage model for agentic commerce: Assistance — AI helps a human research. Selection — AI recommends or selects an option. Transaction — AI executes an explicitly authorized transaction. Delegation — AI receives standing purchasing authority within constraints. Autonomy — AI independently determines when economic action is required. Machine Economy — agents continuously discover, negotiate, buy, sell, and settle with other agents and systems. In August 2026, the overall market is at Stage 3: Transaction. Stages 1 and 2 are mature. Conversational discovery and AI-assisted selection are already normal product capabilities. Stage 3 is real. Native checkout exists in selected ChatGPT and Google integrations. Payment credentials can be constrained. Machine-payment protocols can charge software for digital resources. Stage 4 exists architecturally, but only in narrow implementations. Stage 5 remains experimental. Stage 6 exists primarily as protocols, demos, and early machine-native payment activity rather than a broad operating economy. That distinction matters because much of the market commentary compresses all six stages into one phrase: “agentic commerce.” That hides where the real technical and strategic bottleneck now sits. OpenAI’s commerce strategy is a good example of why the market needs more precise language. Instant Checkout launched in September 2025 with Etsy merchants through the Agentic Commerce Protocol, developed with Stripe. Users could discover a product, approve the purchase, and complete the transaction through ChatGPT while the merchant remained merchant of record. By March 2026, OpenAI expanded ACP more heavily into product discovery and increasingly emphasized merchant-controlled or app-based checkout experiences. That shift has often been described as OpenAI abandoning checkout. That is inaccurate. Native checkout still exists in selected integrations. Instacart, for example, supports browsing, cart creation, and checkout inside ChatGPT. The more accurate conclusion is: