ESW’s new Agentic Commerce platform is arriving with Microsoft Copilot as its first planned integration, putting product discovery, checkout, and payments inside AI shopping experiences rather than sending customers back through a conventional storefront. For merchants running Windows-based retail, fulfillment, and back-office systems, the immediate issue is less about another chatbot integration than whether catalog, payment, and order data can be exposed safely to an AI-driven purchase flow.
Practical Ecommerce’s July 29 roundup identifies ESW, Shoppable, Fastly, Novi, and others as part of a rapid push to turn generative-AI assistants into transaction surfaces. ESW says its platform works alongside existing ecommerce infrastructure, while its July 21 announcement describes Copilot as the initial integration rather than a replacement storefront.
Shoppable is taking a similar route with a Universal Checkout plugin for ChatGPT, based on its Model Context Protocol server. The pitch is a multi-retailer cart and centralized transaction handling, giving brands one connection rather than a bespoke checkout integration for each AI platform.
The timing matters for Windows users because OpenAI renamed its App Directory to the Plugin Directory on July 9, 2026. ChatGPT plugins can package apps, skills, and templates, while workspace administrators control installation and underlying app permissions. That makes commerce plugins an IT governance concern as well as a merchandising experiment: a purchasing action must have clear user approval, scoped access, and an auditable handoff to the merchant’s existing order systems.
Fastly’s partnership with Experian’s Agent Trust ecosystem points to the missing layer. The companies say they will help enterprises assess an agent’s identity, delegated authority, intent, and payment credentials before traffic reaches backend systems. In other words, an AI assistant that can assemble a cart is useful; one that can prove who authorized the spend is essential.
Miva 26R2, released July 21, takes a related approach from the storefront side. Its Vexture merchandising update lets merchants create showcases from AI prompts, searches, categories, or collections, then apply boosting rules to control which products appear first. The release also improves page-cache update behavior and adds multi-select Category Facets, useful improvements for stores where catalog changes otherwise collide with performance and discoverability.
These features will not make bad catalog data good. They do, however, shift more merchandising work from manually maintained page logic toward rules and models that adapt to inventory, query intent, and shopper behavior.
That is the practical limit on the agentic-commerce narrative. A Copilot or ChatGPT purchase experience only improves conversion if inventory is accurate, fraud controls can distinguish delegated transactions from abuse, and fulfillment software receives clean, timely order data.
Merchants experimenting with these tools should treat catalog access, checkout authority, and order-system integration as one project. The flashy interface may be an AI assistant, but the success or failure will still be determined by the data and operations behind it.
Practical Ecommerce’s July 29 roundup identifies ESW, Shoppable, Fastly, Novi, and others as part of a rapid push to turn generative-AI assistants into transaction surfaces. ESW says its platform works alongside existing ecommerce infrastructure, while its July 21 announcement describes Copilot as the initial integration rather than a replacement storefront.
Checkout Is Moving Into the Assistant
Shoppable is taking a similar route with a Universal Checkout plugin for ChatGPT, based on its Model Context Protocol server. The pitch is a multi-retailer cart and centralized transaction handling, giving brands one connection rather than a bespoke checkout integration for each AI platform.The timing matters for Windows users because OpenAI renamed its App Directory to the Plugin Directory on July 9, 2026. ChatGPT plugins can package apps, skills, and templates, while workspace administrators control installation and underlying app permissions. That makes commerce plugins an IT governance concern as well as a merchandising experiment: a purchasing action must have clear user approval, scoped access, and an auditable handoff to the merchant’s existing order systems.
Fastly’s partnership with Experian’s Agent Trust ecosystem points to the missing layer. The companies say they will help enterprises assess an agent’s identity, delegated authority, intent, and payment credentials before traffic reaches backend systems. In other words, an AI assistant that can assemble a cart is useful; one that can prove who authorized the spend is essential.
Search and Merchandising Are Being Reworked Too
Not every change requires an in-chat checkout. Algolia’s Dynamic Facets, released July 21, automatically changes the search filters shown to shoppers based on the query context and behavior. Instead of presenting the same static set of facets for every search, a retailer can prioritize the filters most likely to help a customer narrow results.Miva 26R2, released July 21, takes a related approach from the storefront side. Its Vexture merchandising update lets merchants create showcases from AI prompts, searches, categories, or collections, then apply boosting rules to control which products appear first. The release also improves page-cache update behavior and adds multi-select Category Facets, useful improvements for stores where catalog changes otherwise collide with performance and discoverability.
These features will not make bad catalog data good. They do, however, shift more merchandising work from manually maintained page logic toward rules and models that adapt to inventory, query intent, and shopper behavior.
The Operational Layer Still Decides Whether It Works
The July release cycle also included Maersk’s partnership with ShippyPro, Flowspace’s fulfillment-transition fund, Cleo’s chargeback-prevention update, and Basic 3PL’s Shopline onboarding integration. The common thread is that AI discovery is being paired with conventional operations: carrier selection, warehouse visibility, returns, payment disputes, and routing.That is the practical limit on the agentic-commerce narrative. A Copilot or ChatGPT purchase experience only improves conversion if inventory is accurate, fraud controls can distinguish delegated transactions from abuse, and fulfillment software receives clean, timely order data.
Merchants experimenting with these tools should treat catalog access, checkout authority, and order-system integration as one project. The flashy interface may be an AI assistant, but the success or failure will still be determined by the data and operations behind it.
References
- Primary source: Practical Ecommerce
Published: 2026-07-29T09:00:38+00:00
Loading…
www.practicalecommerce.com