La Banca de Quinielas manages, administers and commercializes the official lottery in Uruguay. It is a business where systems run every day for the whole country, where reliability is not negotiable, and where the software behind the operation has been evolving for decades. In 2026, La Banca decided that AI-assisted development would become part of how its engineering teams work, and chose to prove it first with a group of its developers before extending it to the rest.
Starting from the decision, not the tool
The question at La Banca was never whether developers should use AI. It was how to bring it into teams that maintain critical, long-lived systems without disrupting what already works.
La Banca began with a decision about ownership. The company evaluated the options and purchased Claude Code licenses for its own development teams, so that the tooling, the accounts and the usage would belong to La Banca from day one. La Banca purchased 30 Claude Code licenses, enough to cover about half of its development team, and selected an initial group of developers to adopt the tool first. The idea was simple: work closely with that group, learn from their experience, and only then define how to extend the tool to the rest of the organization. To run that program, La Banca partnered with Qubika, which designed and delivered the adoption process described below.
Adoption in stages, at the teams’ pace
The initial group now works with Claude Code in its day-to-day tasks. The rollout was organized around that group’s real work rather than around a company-wide launch.
The first stage was about setup. The group got its projects configured so that Claude could work within the codebase’s conventions, and built a small set of reusable skills for the tasks the team repeats every day: analyzing tickets, generating unit tests and writing merge requests. That first pass also gave the group a working picture of what Claude Code is and how it fits into their routine.
The second stage came only after the developers had been using the tool for a couple of weeks. At that point, with real questions in hand, Qubika gave the group a practical guide and a session focused on the topic that matters most once the novelty wears off: how to manage context so that Claude works with the right information about the codebase. The group’s feedback was that the content landed better precisely because it came after they had worked with Claude Code for a while, when they could relate it to their own experience.
From there, adoption continued through ongoing support from Qubika as questions came up in practice, along with sessions on best practices and tips, with material adapted into Spanish from Anthropic’s own courses so that the content matched how the group actually works. More recently, the group was given a guide for using Claude Code to prototype new features, extending its use from maintaining existing code to shaping what comes next.
What the developers say
All of the results and feedback in this section come from that initial group, which has now completed the adoption program. The effect showed up in delivery. A project that the group had estimated at six weeks under its previous way of working was completed in three.
The team also tracked how the change felt from the inside, through a short internal survey scored from 1 to 5. Every developer who answered gave the top score to two statements: that the time they spend on repetitive or mechanical tasks, such as writing tests, describing merge requests or searching for context, went down; and that they now use Claude Code in most of the tickets they work on. The statement “I feel I deliver more in the same amount of time” averaged 4.75 out of 5.
The comments were just as clear. One developer described it this way: “It lets us focus on what to do instead of struggling with the technology over how we do it.” Another called it “a paradigm shift in our daily way of working”, comparable to moving from a plain text editor to a full development environment. A third highlighted something the team did not expect: that Claude itself helps them use it better, by creating skills and streamlining processes as they go. One developer working in a different stack from the rest of the team even built a new testing skill for their own environment, using the team’s existing one as a starting point.
From development to the business processes around it
With AI already part of the development workflow, La Banca extended the same lens to its business processes. The company analyzed where AI agents could support existing operations and is now designing and developing those agents to fit into the processes as they run today, rather than rebuilding processes around the technology. On the development side, with the initial group’s adoption complete, La Banca is now evaluating what to add next: which teams follow, and what new uses of Claude Code the group’s experience makes possible.
That sequence matters. The engineering teams first learned how to work with AI on their own code, and that experience now informs how agents are being designed for the wider organization.
Why it matters
Many organizations that run critical systems hesitate to introduce AI into development because the risk of disruption feels higher than the potential gain. La Banca’s approach shows another path: own the tooling, start with a group of developers and their real work, let practice come before theory, learn from that group, and expand only once the practice is in place. It is a model that is careful by design and still moves forward.
