Retail has entered the era of Artificial Intelligence once and for all, but has not yet built the necessary foundation to sustain this transformation. The combination of new generative models, advanced automations and the emergence of so-called “super agents” has accelerated the race for efficiency and personalization. However, this advancement has been happening at a much greater pace than companies' ability to organize governance, security and architecture. It is this discrepancy that worries Marcos Oliveira Pinto, Global Software Engineering Manager at Jitterbit, who directly monitors e-commerce operations at large retailers in Brazil and Latin America.
For him, the sector is experiencing a moment of enthusiasm, but not maturity. “We are seeing many companies trying to get ahead, but leaving essential parts behind,” he says. Although companies with structures enterprise are a little more prepared, most retail still operates on the “do first, organize later” logic. This path, according to Marcos, has a short deadline to take its toll. "I would say somewhere between 12 and 18 months. I believe that, certainly, in 12 months we will start to face a problem with the management and security of this", he exclaims.
And Marcos' warning is not abstract. With the ease of creating AI agents within business areas, and not just in IT, companies are beginning to accumulate automations without traceability, without documentation and without control. Different teams create agents that are disconnected from each other, without standards and without technical supervision. Before long, no one knows how many agents there are, what data they access, how they interact with each other or what risks they pose. The situation is reminiscent of the boom in improvised integrations in the early days of iPaaS, when cluttered environments became expensive and difficult to sustain. Now, this scenario is repeated on a much larger scale, enhanced by the autonomy brought by AI.
The speed of technological evolution also exacerbates the problem and this means that companies may be operating agents with outdated implementations and techniques without even realizing it. Furthermore, the threat of attacks known as prompt injection, where criminals send malicious prompts with the intention of causing unintended behavior and forcing unwanted actions within critical retail operations.
Despite the risks, the potential for AI to transform retail operations is enormous. Marcos cites, for example, situations in which an e-commerce faces instabilities at checkout. Instead of losing sales, a super agent can analyze the customer's history, authorize the purchase to continue and process the data later, avoiding disruptions and reputational damage. In another scenario, an agent can absorb orders while the internal system is offline, ensuring continuity of operation and preventing momentary failures from turning into a drop in revenue.
There are also consolidated applications in dynamic pricing and competition monitoring, in which agents analyze the market in real time and suggest adjustments that keep brands competitive. And, in customer service, agents specialized in sentiment analysis now make it possible to identify customers' moods, frustrations and expectations, paving the way for preventive actions in the customer experience.
To avoid risks and operate with guaranteed security and efficiency, Marcos says that Jitterbit has been investing in super-agent architectures capable of orchestrating different specialist agents — from inventory, logistics, finance, marketing or SAC — within a single intelligent and secure environment. The response reaches the end user in an integrated way, without them realizing which agent is operating. According to Marcos, no other company delivers this approach on the market today, which accelerates the adoption of complex use cases and reduces technical dependencies. When it comes to constantly updating technology, the global company, which also operates in Brazil, has a group of engineers focused on AI, who are constantly testing tools and new solutions.
Even so, he reinforces that technology does not replace maturity. Part of retail's strategic delay is the difficulty in clearly defining which problem it wants to solve. Therefore, the adoption of quick wins, which solve specific pain points, deliver rapid value, and help companies gain clarity on where to scale AI for more complex use cases. At the same time, there is a need for new roles in the market, such as the “AI validator”, a professional responsible for reviewing critical decisions made by agents, preventing hallucinations and ensuring that systems operate within company policies.
The accelerated adoption of AI is not an optional choice, but rather a competitive urgency, as the expert warns. However, a lack of governance and planning can turn this race into a significant operational liability. The sector has, in Marcos' words, up to 18 months to structure its base, before the overload of uncontrolled agents, outdated technologies and security risks become visible and expensive problems. Maturity, therefore, is not just the next step, but the fundamental condition for AI to deliver the value it promises.
