Loading page…
The supplier-published account pairs higher software output with human oversight—and describes plans to combine specialist engineering jobs into one builder role.
Listen to this story
An Anthropic-published case study portrays Mercado Libre’s Claude deployment as an operating-model shift: agents handle maintenance and code review across a 40,000-microservice estate, while finance and other teams use Claude Cowork. Mercado Libre reports sharply higher engineering throughput and a shorter monthly close, but these are customer results published by its AI supplier, not an independent assessment. Its plan for 90% autonomous coding—including pull-request review—and a combined “builder” engineering role remains a target, not a reported current state.
Over the past year, agents working with human oversight reviewed more than 500,000 code submissions and migrated roughly 9,000 services.
In Q2 2026, Mercado Libre reported code submissions up 110%, merged code up more than 100%, and deployments up nearly 75% year over year.
About 40% of code shipped to production in 2025 was written by Claude and reviewed by an engineer; the case study gives no current percentage.
More than 23,000 Mercado Libre developers use Claude Code, and AI agents have reviewed over half a million code submissions with human oversight. A new Anthropic case study describes the deployment’s scale and claimed results, alongside a more ambitious next step: Mercado Libre wants 90% of its coding to run autonomously.
The account also describes Claude moving beyond engineering into finance, procurement and other business functions. Its adoption figures and performance gains are customer-reported results published by Mercado Libre’s AI supplier, rather than an independent evaluation. The distinction matters when reading both the software-output figures and the reported reductions in finance work.
Mercado Libre adopted Claude Code in 2025. It wanted to shift infrastructure upkeep onto the tool, including software-version migrations, dependency patches and test coverage across 40,000 microservices—the smaller services that make up its systems. The deployment covers both individual developers’ work and larger automated tasks, including code review and generating tests.
Over the past year, agents working with human oversight reviewed more than 500,000 code submissions and migrated roughly 9,000 services to newer platforms. That makes the deployment more than a tool for suggesting code to individual engineers: the company is also using agents to carry out maintenance across its software estate.
Mercado Libre reports code submissions rose 110% year over year in Q2 2026.
The company says merged code increased more than 100% over the same period.
Deployments rose nearly 75% year over year, according to the case study.
Mercado Libre also reports fewer rollbacks—reversals of software changes—year over year, though it gives no numerical reduction. In 2025, roughly 40% of code shipped to production was written by Claude and reviewed by a human engineer before release. The case study says human-written code had become the exception by 2026, without supplying a current percentage.
Finance was the first non-engineering team to adopt Claude Cowork. Within a month, most of that team was using it. Shared skills helped employees generate branded presentations in minutes, replacing work that previously took several people hours each week. The larger process change concerned the monthly financial close: Mercado Libre says Cowork shortened it by a third.
The finance team can now preview the close on demand instead of waiting until month end. Claude also drafts initial explanations for changes in results using Mercado Libre’s business context. Business and finance teams then validate those explanations. Here, the described workflow keeps people checking the interpretation, even as the software prepares more of the underlying work.
Cowork subsequently spread to marketing, HR, legal, compliance, and business and commercial teams. The account puts its reach across those groups at more than 26,500 people, with 78% active weekly. Some procurement teams built Claude-powered agents to answer supplier questions without involving IT, extending automation beyond centrally managed engineering projects.
The shift has already brought some managers back into hands-on development: the case study says 30% of managerial roles now contribute code again. But the next steps remain goals. Mercado Libre’s autonomy target includes the full pull-request review process—the checking of proposed code changes—not just writing the code itself.
Mercado Libre also plans to combine separate backend, frontend and database roles into a single builder role, with agents filling the gaps. That proposed structure would change how engineering responsibilities are divided, rather than simply giving each existing specialist a faster coding tool.
Loading discussion...
Join the conversation
Explain where broader responsibility would help—or where specialist expertise should stay.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.