AI Agents are here but systems can’t speak the language. A new generation of autonomous AI agents is ready to automate complex commerce worklfows. They can reason, plan and execute tasks however their potential is capped by the fragmented API landscape. Each integration is a bespoke connection limiting the scale and power of AI-drive automation. That is set to change..
What does it take for an AI to run an entire digital advertising campaign, manage the logistics, and handle the returns?
We are not there yet but what we are witnessing is the assembly of a new stack that allows AI to plug directly into the engine of global commerce. The new world of agentic AI is being built on a foundational standard called the Model Context Protocol (MCP) which standardises how AI applications communicate with external systems solving the fragmentation problem that has traditionally plagued the development of end-to-end systems. MCP enables systems to talk in human language moving from tightly coupled integration where system A has to know exactly how to talk to system B.
A concrete example of this is the new “Instant Checkout in ChatGPT” feature. Users can now move from a conversational query—like “help me find a handmade, ceramic dinnerware set, in white and tan under $100″—to a completed purchase of a “Dinnerware Set by BlancPottery” on Etsy, all within a single conversation. OpenAI and Stripe came up with ACP (Agentic Commerce Protocol) which is an open standard, that allows an AI agent to securely manage the entire checkout process. It provides a common language for coordinating the purchase and securely sharing payment credentials between buyers, agents, and businesses.
There are two other protocols that sit pre order and post order
Firstly AdCP (Advertising Control Protocol) which is a unified interface for managing the entire advertising lifecycle. It solves the core fragmentation issues in advertising the Integration Problem (countless APIs), the Discovery Problem (scattered inventory across multiple systems), and the Automation Problem (no standard for AI agents). AdCP allows an agent to discover audiences, compare media, and launch campaigns across any platform using a single language.
Then there is onX (Order Network eXchange) which is a specification designed for all the complex “back office” work that happens after a purchase. It provides a standard way for an AI to place orders with fulfillment centers, track inventory levels, manage shipments, and handle returns. It creates a standard way for an AI to communicate with the sprawling, fragmented web of Order Management Systems (OMS), third-party logistics warehouses (3PLs), and Enterprise Resource Planning (ERP) systems that run the global supply chain.
Perhaps the most surprising aspect of this new infrastructure is that it’s being built collaboratively. In a departure from previous tech eras defined by proprietary standards and walled gardens, the foundational protocols for agentic AI are being developed as open, cooperative standards.
