How agentic AI is changing retail from within

Retail has grown accustomed to conversations about artificial intelligence in the context of recommendation algorithms and support chatbots. But the most serious changes are happening where no customer ever looks — in engineering departments, where code is written that controls all of retail's infrastructure. This is precisely what Prasad Banala, director of software development at one of the largest retail operators in the United States, discussed in a podcast from the Infosys Knowledge Institute, published by MIT Technology Review.

Banala describes an approach that goes far beyond the conventional use of AI as a programming assistant. His team has deployed agentic AI — systems capable of autonomously performing complex multi-stage tasks — at all stages of the software development lifecycle. This begins with requirements validation: AI agents analyze business requirements, identify contradictions and gaps before engineers write a single line of code. Then agents participate in code generation and test automation, closing a cycle that previously required constant manual oversight.

It's important to understand the context here. The term "agentic AI" has transformed over the past year from an academic concept into the leading trend in corporate technology. Unlike ordinary language models, which respond to requests on a "question-answer" basis, agent systems are capable of planning actions, using tools, interacting with external services, and adjusting their strategy based on intermediate results. Essentially, this is a shift from AI as an advisor to AI as an executor. And retail, with its enormous data volumes, complex logistics, and constant margin pressure, proved to be an ideal environment for such a shift.

However, the most interesting aspect of Banala's account is not the technological side, but the organizational one. According to him, the main barrier to functioning agentic AI is not model quality or computational power, but teams' readiness to restructure their processes. When an AI agent takes over requirements validation, the role of a business analyst doesn't disappear, but undergoes radical transformation. The specialist ceases to be someone who manually checks documentation and becomes someone who sets rules for the agent and controls the quality of its outputs. A similar transformation occurs with testers, architects, and project managers.

This shift has serious consequences for the entire industry. Retail is one of the world's largest employers, and its technology divisions employ thousands of engineers. If the model described by Banala proves its effectiveness at scale, it will inevitably spread to other sectors: banking, insurance, telecommunications — anywhere large development teams with established processes exist. In fact, major retailers are becoming a testing ground for corporate agentic AI, and the results of these experiments will determine how engineering organizations will be structured in three to five years.

It's worth noting the risks as well. Autonomous AI agents in the development cycle are not the same as a chatbot that suggests code. An agent error at the requirements validation stage can cascade throughout the entire project. Questions of responsibility, audit, and quality control in such systems don't yet have standard solutions. Banala mentions that his team builds multi-level verification mechanisms, but few details are provided — which is unsurprising for a company working in a competitive environment.

The experience of a major American retailer demonstrates an important pattern: the future of agentic AI is decided not in laboratories and not at conferences, but in the everyday engineering practices of large organizations. It is there, in the collision of ambitious technologies with real processes, that models are born that will either transform the corporate world or remain costly experiments. For now, apparently, retail is betting on the first scenario.