Artificial Intelligence (AI) has been a force in the retail industry for nearly half a century.  While there is much data to show the widespread adoption, there is just as much that indicates the disappointing results for many.  As the industry now enters the era of Agentic AI, there is a huge opportunity to use AI in a networked, coordinated fashion to make business processes faster, smarter and often cheaper. 

Retail strategist and tech industry veteran Jon Stine has spent decades working with retailers at the intersection of business performance and technology.  In this article he presents insights into how retailers should be thinking about applying Agentic AI to their business strategy to move into retail’s future – no matter where they are on their journey. 

Jon Stine is a Founding Fellow of the Honeycomb Retail AI Collective.

The retail industry’s future will be driven by agentic artificial intelligence.  

The industry’s first AI phase – one that began roughly fifty years ago and stretched to the first decades of this century – used predictive machine learning for such things as market basket analysis, demand forecasting, inventory management, and product recommendations based upon a shopper’s browsing and purchase history.  

A second phase erupted with OpenAI’s November 2022 launch of ChatGPT 3.5.  Since then, generative artificial intelligence – be it in text, voice, image, or video – has become, across industry events and C-suites, the operative definition of AI. 

Generative AI has won widespread adoption. A March 2025 study showed that more than 78% of companies are now using generative AI in at least one business function. 

But the same study showed that more than 80% of the adopters reported no material contribution to earnings from their gen AI investments. And a July 2025 study from MIT found that 95% of organizations deploying generative AI were getting zero returns.  

We shouldn’t be surprised.  

Generative AI has largely been adopted for individual, siloed productivity. It has saved time and sparked awe but rarely touched business processes or workflows – the realms in which P&Ls are shaped.  

Figure 1: the crossing of the AI Value Chasm. The tech visionaries focused solely on generative AI will find themselves trapped; those with a pragmatic, business-first, multi-AI approach (led by agentic AI) will have a much higher probability of crossing the chasm and joining the industry’s AI leaders in transformative value creation. 

WELCOME TO THE THIRD ERA OF AI FOR RETAIL: AGENTIC ARTIFICIAL INTELLIGENCE. 

The new, emerging third era of AI for retail is agentic artificial intelligence. 

Agentic AI is artificial intelligence agents working with other AI agents – in a networked, coordinated fashion to make a business process faster, smarter, and/or cheaper.  

Agents are small bits of AI that do things. They’re digital task rabbits. They take in data, analyze data, and act upon the analysis.  

Agents can be limited in scope (the management of building temperature and light) or combined in increasingly complex ways to automate, speed, and smarten a workflow. (A self-driving car is an orchestra of agents with wheels.) 

Value is multiplied when agents send information and insights to other agents within and across business processes. The output of one becomes the input of another, and then another – informing, advising, harmonizing, even across operational areas. (Say, from marketing to merchandising to operations.)    

THIS IS ARTIFICIAL INTELLIGENCE BUILT TO CREATE BUSINESS VALUE.  

Envision multiple agents working together to automate/manage/optimize a complex workflow. Such as the chain that stretches from vendor raw materials to distribution centers and store shelves.  

Envision data, knowledge, and insight that move automatically up and down decision levels. Think of agents of distinct functions and capabilities – some gathering data, some analyzing it, some sending messages. Think about agents that serve as managers, as “orchestrators,” of an agentic system. Think of agents that serve as monitors of accuracy or manage data access. 

Envision bees in a hive. 

WHAT’S THE VALUE MATH OF AGENTIC AI? 

Agentic artificial intelligence, when brought into retail workflows, will create marginal gains. Slow starts, big finishes – at least for those with foresight and patience. A little faster here, a little smarter there, a little less expensive over there. Repeat. Reuse data in a next-door workflow and bring in new insights. The marginal gains will, in time, multiply exponentially.  

INDUSTRY LEADERS ARE DOING AGENTIC. NOW.  

It will come as no surprise that the industry’s leaders – from brands in places such as Bentonville, London, Neckarsulm, North Minneapolis, Stockholm, and Ingolstadt – are now testing and implementing agentic artificial intelligence solutions.  

Use cases range from the analysis of multiple areas of customer behavior (from traffic to transaction) to streamline store layouts and lift conversion, to contact centers, marketing and messaging automation, delivery process optimization, and multi-channel dynamic pricing. Several large grocery chains are reported to have used orchestrated multi-agent systems during heat waves to monitor social signals and inventory, launching timely promotions and reallocating stocks. 

WILL AGENTS REPLACE ME?  

Agentic AI should not be about headcount reduction. This is not a future of bots making autonomous merchandising decisions, nor of human-less retail. In the eternal question of employee augmentation vs. AI-driven staff attrition, leaders are repeatedly choosing augmentation – using agentic AI to free employees from mundane tasks and bringing broader-deeper intelligence to human decision-makers.  

WHAT ARE THE OTHER ISSUES? 

Agentic AI will require a business-first development approach. Technology is the last issue to resolve. The best implementations will start with an assessment of brand pain points and competitive positions and work its way to relevant workflows and data sources – and then assess the relevance of AI capabilities. Should the discussion begin with a comparison of LLM’s, stop and start over.  

Agentic AI will – like every digital project – require clean, accessible, and smartly-tagged data. Agentic AI will use data from inside and outside the firewall, data that is both structured and unstructured.  

Agentic AI will require standards and protocols for interoperability, so that agents can access the data they need and actions can be shared. Most enterprise agentic environments will grow in a piecemeal, heterogenous way over time.  

Agentic AI will demand decisions on data access, decision rights, and system transparency. Which agents (on behalf of which humans) should have access to what data? Are agents opaque boxes of algorithmic mystery, or can it be cracked open to see why it’s deciding what it’s deciding?  

Agentic AI will demand accuracy at every level. In an agentic system, nonsense will amplify. Developers are now working to create agentic monitors of veracity.  

THIS IS OUR FUTURE OF DOING RETAIL.