Mercado Libre isn’t just tinkering with AI anymore. The Latin American e-commerce giant has deployed Verdi, a new internal development platform built on OpenAI’s GPT-4o, that’s already autonomously resolving customer disputes involving serious money — $450 million annually, to be specific. That’s a massive leap from the AI-assisted product tagging and review summaries the company has been using, and COO Daniel Rabinovich didn’t mince words about the difference. “Using AI for cataloging and summarizing reviews is interesting and helpful, but not transformational,” he said. Verdi, he argues, flips the script entirely.
The platform isn’t a simple chatbot. It’s a neural-network-like framework that lets Mercado Libre’s 17,000 developers string together AI skills, Python nodes, and internal APIs without ever touching source code. A “return” skill can autonomously negotiate with a “shipping” skill, for instance. This architecture is currently handling 10% of mediation on a major site, and the company sees a path for Verdi to eventually support the workload of its 9,000 human customer service agents. The AI doesn’t just make decisions; it reaches out to other internal systems and even humans when guardrails require it, making the human operator a “co-pilot to the AI,” as Rabinovich put it.
SVP of Technology Sebastian Barrios explained that the company couldn’t find an existing platform that balanced rapid iteration with the necessary security and scalability. So they built their own on GPT-4. The emphasis on natural language as the central development interface is a deliberate choice to lower the cognitive load for their massive team managing over 30,000 microservices. This isn’t just about automating simple queries; the system uses GPT-4o’s vision capabilities to process “proof of damage” photos in disputes, interpreting what’s broken or missing from a returned item.
The scale of what’s already automated is staggering. Beyond the $450M in disputes, Mercado Libre credits OpenAI’s models with a 100x increase in cataloged products over two years and a fraud detection rate nearing 99% on flagged listings. But Verdi represents a more profound shift. The company is wiring autonomous decision-making directly into high-stakes financial operations. The real test will be what happens when those automated decisions go wrong and the “co-pilot” human has to take back the controls they’ve been trained to hand over.
