Every retail CTO I meet asks me a version of the same question.

“You're the third AI merchandising vendor we've evaluated this quarter. Each one promises to be the central nervous system of our merchandising operation. We already have Oracle Retail. We have SAP. We have Blue Yonder for some categories. We're piloting Salesforce Agentforce. How is your AI going to talk to all of that without us building integrations forever?”

It's a fair question. And for most vendors, the honest answer is: it isn't.

That's why we built Intelo differently. Not as another platform that wants to own your stack, but as connective tissue — interoperable merchandising agents that plug into the systems and orchestration layers you already have, using the open standards that the rest of the agent ecosystem is converging on.

If you're the CTO or CIO sponsoring AI initiatives in retail, the next two sections are the most important thing you'll read this week.

The new shape of the retail AI stack

Your stack is not a single platform anymore. It is a layered ecosystem.

At the bottom are your systems of record — ERP, PLM, warehouse management, POS, planning tools. These have ten- and twenty-year tenure. They're not going anywhere.

In the middle, a new layer is emerging: enterprise agent orchestrators. Salesforce Agentforce, Microsoft Agent Framework, ServiceNow's agentic platform, Publicis Sapient's Agentic Retail Network. These are control planes — they route work between agents, manage identity, log decisions, and present a coherent surface to your business users.

At the top sits a growing set of specialized, domain-deep agents — for merchandising, logistics, marketing, customer service, finance. These are the workers.

The hard question for any CTO is: are the agents I'm bringing in going to be good citizens of this stack? Or am I buying intelligence today that becomes a silo tomorrow?

Two protocols are settling the question

The industry has, mercifully quickly, converged on two open protocols that answer how agents should connect.

MCP — Model Context Protocol. Originally proposed by Anthropic, now broadly adopted, MCP is the USB-C port of AI. It is a standard way for any AI agent to securely access an enterprise data system or tool — without a bespoke ETL job, without a fragile screen-scraper, without a vendor-specific connector. You expose your data system once via MCP, and any compliant agent can use it.

A2A — Agent-to-Agent. A2A is the universal language for agents to discover, delegate to, and collaborate with each other. An A2A-native agent publishes an AgentCard describing what it knows how to do and how to be invoked. An orchestrator — or another agent — can discover it, route a task to it, stream the result back, and pass control on. No screen-scraping. No human-in-the-middle copy-paste between dashboards.

If your AI vendor isn't speaking both of these, they're building you a silo, whether they admit it or not.

How Intelo is built

Intelo's agents — across Merchant Analytics, Merchandise Financial Planning, Assortment Planning, In-Season Management, and Pricing & Promotion — are A2A- and MCP-native from the ground up. Not retrofit. Not roadmap. Production.

Two architectural choices make this real.

First, our MCP gateway sits in front of all enterprise data access. When the Open-to-Buy Agent needs to read your latest commitment positions, or the Replenishment Agent needs current store-level inventory, it doesn't call a custom Intelo connector — it calls MCP. We've already done the work of connecting MCP to Oracle Retail, SAP, Microsoft Dynamics, Salesforce, Fluent Commerce, Blue Yonder, and Manhattan Associates. When you bring in a new data source — or want to swap one out — you don't wait for a vendor release cycle.

Second, our A2A agent network. Each Intelo agent is a first-class A2A citizen. Every agent publishes an AgentCard with its skills, its inputs and outputs, and its invocation contract. Our SuperAgent uses A2A to route between Intelo's own agents — and just as importantly, it can route to and from agents that aren't Intelo's. When a customer's orchestration layer sends a task into the Intelo network, that task is just an A2A message. When the Markdown Strategy Agent needs to hand a recommendation back to the customer's enterprise approval workflow, that too is just A2A.

We are not the only specialized agent you will run. We don't want to be.

What this unlocks for the CTO/CIO

Three things matter here, and they are all about optionality preserved.

You keep your orchestrator. Whether you've standardized on Agentforce, Microsoft Agent Framework, ServiceNow, or your own internal platform — Intelo plugs in. Our agents are exposed via A2A endpoints that any compliant orchestrator can call. We have shipped partnerships with Microsoft Agent Framework, Salesforce Agent Orchestration, and Sapient's Agentic Retail Network to prove this isn't theoretical.

You don't pay an integration tax. Custom data pipelines die hard. They are the single biggest reason enterprise AI projects miss their go-live. By standardizing on MCP, Intelo eliminates the bespoke integration layer. We have Balenciaga, The Children's Place, and Psycho Bunny live with production data flows. The expensive part — engineering custom connectors for every data source — is gone.

Your specialists can be specialists. The dirty secret of the “all-in-one platform” pitch is that no team can stay world-class at fifteen different domains. Merchandising is hard. Logistics is hard. Customer service is hard. The architecture you actually want is best-of-breed agents collaborating cleanly — and that only works if the protocols are open.

A bet on open standards

When my co-founders and I started Intelo, we made a deliberate choice that some advisors told us was risky: build for interoperability first, even if it slowed our initial product surface.

Eighteen months later, that bet is paying off. Customers don't ask us to replace their stack anymore. They ask us to make their stack intelligent. Orchestration partners don't see us as competition — they see us as the domain depth their platforms need. And the open-standards thesis has moved from contrarian to consensus faster than even I expected.

To CTOs and CIOs evaluating merchandising AI: ask every vendor in your pipeline two questions.

1. Are you A2A-native?
2. Are you MCP-native?

If the answer is anything other than yes — anything about “we have an SDK,” anything about “we're on the roadmap,” anything about “we can build a custom integration” — you are being sold a silo.

Merchandising AI is going to be transformational for retail. But it is only going to be transformational at the level of the whole enterprise, not as another isolated tool. That requires good citizens. We built Intelo to be one of them.

If you want to see what a radically interoperable agent network looks like running against your data, in your orchestration layer, we'd be happy to show you.