MCP

The Protocols Powering Agentic Commerce: A September 2026 Update

by

Where things stand

A year after the first agent checkout went live, the protocols behind agentic commerce have settled into clear layers, most have moved to neutral governance, and real agent-led spending is still small. Our earlier article on the protocols powering agentic commerce laid out the standards that were starting to let AI agents shop and pay on our behalf. Since then, several of those protocols have changed owners, changed scope, or outlived the products they were first built for.

How the layers fit together

The protocols are best understood as layers that work together rather than rivals. Each one solves a different part of the same problem, from letting an agent talk to a store to proving a person approved the purchase.

Read more

MCP vs. RAG: Two Ways AI Gets Smarter

by

Your AI chatbot is only as smart as what it knows and how it’s allowed to act on that knowledge. That distinction is at the heart of two of the most important technologies shaping enterprise AI right now: Retrieval Augmented Generation (RAG) and the Model Context Protocol (MCP).

If you’re investing in AI tools, piloting copilots, or signing off on AI roadmaps, understanding the difference isn’t a technical detail, it is a critical one.

Large language models (LLMs) like the ones powering ChatGPT, Claude, or Copilot have a fundamental limitation: their knowledge is frozen at the time they were trained. Ask about last quarter’s sales figures, your internal HR policy, or a new regulation published last month, and without additional help, the model either guesses or says it doesn’t know.

RAG and MCP are two different answers to that problem. One improves what the AI knows. The other improves what the AI can do.

Read more