4 min read

What I learned from Shoptalk Fall: Agentic commerce still depends on integration

Q: What is agentic commerce?

Agentic commerce is a model where an AI agent interprets shopper intent and directs the customer to a retailer to complete the purchase. The retailer keeps the sale and must supply product data accurate and current enough for a machine to read.

Q: Why does AI in commerce depend on integration?

AI agents act on the data they can see. If inventory, pricing, and margin data live in disconnected systems, the agent acts on an incomplete picture. Automation can run without being connected, and that is how errors scale.

Q: How should IT teams govern MCP connections to business systems?

Define what each agent can read and write, require approval for sensitive actions, and keep an audit trail. These are the same integration controls already used for other systems.

Q: How should a company roll out AI agents in operations?

Validate the data, reproduce real business logic, keep a human approving while the system learns, then expand autonomy.

Q: What does Celigo’s Product Launch Readiness agent do?

It reads the real state of every connected system before a launch goes live, so issues like a bad price surface as a question for a human days before launch.

Published Oct 7, 2026

Artie Sharpe

Senior Ecommerce Director

I spent twenty years as a commerce operator before joining Celigo. I went to Shoptalk Fall looking for what had changed. The front end has. The back-end problems are the same ones I had before, and they are why many brands struggle to make AI work the way they want.

Key takeaways for IT leaders

Agentic commerce conversion is small today. The work to do now is making product data accurate and current enough for a machine to read.
Customers asking about MCP all asked the same thing: can we govern it? That is an integration question.
The agents paying off today sit inside the operation, not in front of the shopper.

What is agentic commerce, and what does it require from IT?

Agentic commerce was everywhere on the Shoptalk Fall agenda. Anthropic laid out a model where Claude reads shopper intent and sends the customer to the retailer to check out. The retailer keeps the sale. In return, the retailer owes product data that is accurate and current enough for a machine to read.

Several sessions made the same point about timing. Conversion through agentic storefronts is small today. The work to do now is getting your data in order. For a VP or Director of IT, that means the product catalog, pricing, and inventory across your commerce platform, ERP, PIM, and marketplaces need to agree with each other and stay current.

How do you govern MCP and AI assistants connected to business systems?

Between sessions, customers wanted to talk about MCP (Model Context Protocol). Most already use an AI assistant every day. Now they want it connected to their own systems, and every one of them asked the same thing: can we govern it?

Governing AI access comes down to four questions:

What can the agent read?
What can it write?
Who approved that?
How would we know?

Those are integration questions. Commerce teams have been answering them for years for every system they already own, through access controls, approval flows, error handling, and audit trails. An AI agent is another consumer of connected systems, and it should meet the same standards.

Where are AI agents delivering results in commerce operations today?

The agents paying off today sit inside the operation, not in front of the shopper. The brands reporting results are:

Sensing demand from store-level data
Reconciling inventory records against what is physically on the shelf
Classifying products for cross-border shipping before an order is created
Checking promotional pricing across systems before it goes live

Each of these depends on connected systems and trusted data. The operations panel gave a sequence worth following:

Validate the data.
Reproduce your real business logic.
Keep a human approving while the system learns.
Then widen autonomy.

What happens when automation runs on disconnected data?

The cautionary tale came from the marketplaces session. A brand ran Black Friday pricing with margin data missing. Products hit the marketplace at the wrong price and generated thousands of orders the brand could not fill. Automation did exactly what it was told.

This is what Celigo’s Product Launch Readiness agent is built for. It reads the real state of every system before a launch goes live, so a bad price shows up as a question for a human two days out instead of a pile of orders.

We’ve built agentic templates to help commerce teams accelerate building their AI workflows. Here are three you can turn on today.

Why integration is a strategy, not a project

None of this is new to anyone running commerce operations. Connected data across your storefronts and systems was always the goal. AI raised the stakes of not hitting it. Things can be automated without being connected, and that is how a pricing run ships thousands of orders nobody can fill.

I continue to believe integration is a strategy, not a project. You do not finish it and move on. You build it so whatever comes next lands on a good foundation. That is the work we do, and after three days of sessions and conversations I am more convinced this is what drives success.

Want to see how Celigo can help you build the integration strategy for agentic commerce? See the platform in action.