By Shyam Rao, Founder and CEO, Tote.ai

If you read my last article in this newsletter, you walked away with a framework for evaluating AI claims in convenience retail. Five questions to ask vendors. Ways to tell whether the intelligence in a system is structural or cosmetic.

This article builds on that foundation. The industry is still sorting out what genuine AI capability looks like at the point of sale. Some vendors are calling faster software development an AI story. Others have added AI to individual products without a coherent architecture underneath it. It’s worth asking hard questions about any of these claims or approaches before taking them at face value. But even setting that aside, there’s a significant gap between AI that surfaces an insight and AI that acts on it, and almost no one in this industry is on the right side of it. Many vendors, if they’re delivering anything, are delivering AI assistants. Agents represent a quantum leap past that. One answers your question. The other changes what’s possible in your operations.

Why the distinction matters at the point of sale

Convenience retail is a high-velocity environment. Transactions are fast. Associates are often new. Customers are in a hurry. The value of AI in this context isn’t just knowing the right answer. It’s reducing the number of steps between a problem and its resolution or addressing potential problems before they ever happen.

Consider a few scenarios.

Store Promotions

An AI assistant can tell a manager what promotions are currently active if they ask. An AI agent sees sales are dropping, analyzes traffic patterns, inventory and margin data, identifies a promotion likely to drive revenue, and either flags it for manager approval or deploys it autonomously, depending on how the operator has configured it.

Maintenance Tickets

A receipt printer jams at Pump 4. An AI assistant can answer questions about pump status if asked. An AI agent creates the service ticket, routes it to the right department and tracks resolution, before a customer has to complain.

These are AI agents applied to retail operations. Significant on their own. But they’re also the foundation for something bigger.

The spectrum operators should understand

It helps to think about AI capability in physical retail as a series of exponential leaps, not a feature list.

Stage 1 | AI Assistants: Retrieving Information on Demand

The first stage is the AI assistant: reactive, retrieves information, answers when asked. Many vendors, if they’re talking about it at all, are at this stage, or claiming to be.

Stage 2 | AI Agents: Taking Action Autonomously

The second stage is the AI agent: autonomous, connected to live systems, empowered to act. This is where the distinction gets important. The agent is the only intelligent entity in the room. The systems around it—the POS, the pump controller, the inventory database—hold data but can’t reason or act on their own. The agent supplies the intelligence on their behalf. It reads pump status, interprets it and does something about it. The power here is access. Very few vendors are genuinely at this stage, and the gap between claiming and delivering is significant.

Stage 3 | Agentic Ecosystems: Multiple Agents, Optimizing Together

The third stage is where the industry is headed, and it’s architecturally different from the second. It’s not one intelligent agent operating inside passive systems. It’s multiple agents, each representing a different party, each optimizing for different outcomes, communicating with each other to find solutions neither could reach alone. A retailer’s inventory agent sees a high-velocity item trending low and reaches out to a supplier’s fulfillment agent to negotiate replenishment—timing, volume, price—without a purchase order or a phone call. The outcome isn’t scripted. It’s emergent.

Agentic commerce is the commercial expression of a consumer’s AI completing a purchase with a retailer’s AI, on behalf of the people they each represent, without a human initiating each step. Physical retail is where this gets most interesting, because the transaction starts at the forecourt, often before a human associate is involved at all.

E-commerce companies have been building toward this for years. The agentic commerce conversation, from Google and Mastercard to OpenAI and Shopify, is happening almost entirely in online retail today. Physical retail hasn’t entered that conversation yet. That’s the opportunity.

The infrastructure question

Getting to agentic commerce isn’t just an AI question. It’s an architecture question.

For an AI agent to take action in your store, it needs access to the same systems your associates use. It needs to read live pump data, write to a cart, apply a promotion, create a ticket. That requires an open, API-first platform where AI operates through the same interfaces as everything else.

If your POS was built 20 years ago and AI has been bolted on as a layer, your agents can’t reach the underlying system. They can describe what’s happening. They can’t change it. And they certainly can’t interact with a consumer’s AI agent on the other side of a transaction.

This is the practical test for any vendor making agentic claims: Can your AI take action through the same APIs your platform exposes to customers? Or does it only have access to a documentation library?

The operators who move now will build advantages that compound. Those who wait won’t just fall behind. They risk locking themselves into infrastructure that can’t participate in this world at all. The window is open today. It won’t stay that way.

What comes next

We’re early in this shift. Most operators are still evaluating whether their current AI vendor can clear the bar from my last article. That’s the right first question.

But the second question is the one that will define competitive positioning over the next few years: Is your technology capable of acting, not just answering? And is it built to participate in a world where your systems need to talk to your customers’ systems directly?

Find Out More on April 16

We’ll be exploring what this looks like in practice in an upcoming SG Voices webinar on April 16 at 2 p.m. Eastern. The focus will be on the forecourt as the entry point for this kind of engagement, what operators who are modernizing are noticing first, and how the path from fuel customer to in-store buyer is getting shorter with the right infrastructure in place.

If you’re thinking seriously about where AI takes your operation next, it’s worth tuning into the webinar. Register here: SG Voices April 16 Webinar

Shyam Rao is the founder and CEO of Tote.ai, an AI-native point-of-sale platform built specifically for fuel and convenience retail. He previously co-founded Punchh, an AI-powered customer engagement platform serving nearly 350 enterprise brands, which was acquired for nearly $600 million.