​From assistants to agents: What agentic AI really means for retail

Agentic AI is the newest trend in retail tech. The idea of AI agents that can plan, execute, and optimise tasks with minimal human input is generating headlines and curiosity in equal measure.

But as with every wave of hype, we need to pause, ground ourselves in the realities of retail, and ask: what’s genuinely useful today and what’s still theoretical? What’s real, what’s not, and how can retail leaders navigate the rise of agentic AI? (Think machine learning and supply chain automation, which delivered real value, versus blockchain or the metaverse, which promised more than they’ve proven.)

Assistant. Agent. Agentic. What’s the difference?

We’re witnessing a compression of the hype cycle. Concepts that once took years to mature now hit headlines before they’ve even been piloted.

Let’s draw some much-needed distinctions:

• AI assistants support specific tasks. You prompt them, they respond. Think chatbots or text generators.
• AI agents manage a set of related tasks to achieve an outcome, such as detecting low stock on a bestselling item, placing a replenishment order and notifying the store team when it’s due to arrive.
• Agentic AI, the shiny new term, refers to systems that can autonomously define goals, break them into subtasks, execute across tools and adapt based on feedback – for example, identifying a drop in category performance, diagnosing the root cause (pricing, availability, competitor action), designing a targeted promotion, coordinating media spend and adjusting the strategy in real time based on sales response.

It’s a compelling vision. But let’s be honest, most so-called ‘agentic’ tools in retail today are simply well-crafted automations with a fresh coat of AI branding. They often lack the deep, domain-specific intelligence needed to navigate the complexity of modern retail.

What AI can do today

There are very real, tangible use cases already delivering business value:

• Replenishment support – monitoring stock data and recommending orders before gaps appear;
• Product copy generation – creating tailored descriptions for thousands of SKUs;
• Personalisation – delivering relevant offers across channels;
• Dynamic pricing and promotions – balancing value perception and margin;
• Assortment optimisation – tailoring ranges by mission, format and location

These applications are active areas of innovation. But the moment we expect AI to run end-to-end decisions in retail without oversight, we risk overshooting what’s actually possible, or safe.

Retail is not a clean optimisation problem. It’s a system of trade-offs, constraints and interdependencies. Price affects demand, which affects inventory, which shapes promotions, which influences media, which in turn affects customer behaviour. Everything is connected and constantly shifting.

Recent coverage of AI agents ordering groceries on behalf of customers highlights how technology can shift customer expectations. While most shoppers aren’t there yet, innovations like this hint at a future where convenience is redefined, and where customers may expect retailers to act on their behalf, not just respond to their requests.

Walmart’s recent AI roadmap captures this well, not just through its vision for customer experiences but in the infrastructure it’s building beneath them. The company describes this shift as “moving from AI features to AI systems”; investing in orchestration layers, governance structures and reusable tools that will enable intelligent agents to operate safely and at scale. It’s a reminder that agentic AI isn’t a single tool, but a system that depends on deep integration, clear logic and real-world constraints.

“Let’s not get lost in the tech. Let’s stay focused on what really matters, which is helping customers, supporting colleagues and building better businesses”

Five actions retail leaders should take

• Think in systems, not features. Agentic AI isn’t plug-and-play. It’s an ecosystem that requires orchestration layers, data infrastructure, and governance. Focus on building capabilities that can scale safely, not just launching isolated pilots.
• Get specific about AI’s role. Don’t adopt agentic AI because it’s trending. Use it where it solves a real, meaningful problem, especially where it frees up colleagues to do higher-value work like customer interaction, strategic thinking and innovation.
• Invest in people, not just platforms. Train colleagues to work alongside AI. Build capability in data literacy and judgement-led decision making.
• Design for trust, not just output. Whether it’s pricing, personalisation or replenishment, customers and colleagues need to understand how AI decisions are made. Transparency, explainability and fairness aren’t optional, they’re essential to adoption.
• Strengthen your science foundations. Agentic systems are only as good as the intelligence behind them. If your current tools struggle to diagnose root causes, autonomy won’t fix that, it will just scale poor decisions. Focus on building robust, retail-specific science that can handle complexity and drive meaningful action.

In summary, agentic AI needs a retail brain

Agentic systems may bring autonomy. But they won’t deliver value unless they’re grounded in the realities of how retail works – complex, constrained and constantly evolving.

Let’s not get lost in the tech. Let’s stay focused on what really matters, which is helping customers, supporting colleagues and building better businesses.