From AI Agent to AI Colleague: the coding revolution

Everyone's talking about it: Artificial Intelligence is transforming the world of work. But beyond the headlines and tools promising miracles, what does it really mean to integrate AI into the processes of a tech company like ours?

In the Backend Solutions area, we decided to find out for ourselves by launching an experimental process. We didn’t just “talk” to the AI—we hired it. Well, almost. We're integrating an AI Agent into our team: a digital colleague working side by side with our developers.

Not just a tool, but a digital colleague

We weren’t looking for an “enhanced prompter” for our developers, but a truly autonomous agent capable of interacting and taking on entire tasks. The idea is as simple as it is powerful: automate routine activities—such as simpler or repetitive tasks like documentation and testing—to free up our developers’ time and energy.

Human talent is too valuable to be spent on such routine tasks. Our medium-term goal is to increasingly focus people’s added value on what truly makes a difference—like system architecture, problem solving, and the implementation of complex software.

Agent Who? On the shoulders of giants

But an agent this advanced doesn’t just appear out of nowhere—especially with technology evolving so rapidly. To ensure the power and reliability we needed, we turned to the best. We formed a strategic partnership with Anthropic to leverage their Claude coding models, specifically designed for the enterprise world. This gives us not only top-tier performance, but also the security and stability our clients’ projects demand.

We're also collaborating with Google to integrate and test the latest versions of their Gemini Coder model, which continues to show increasingly promising performance.

But how does it actually work?

What’s most interesting is that we didn’t have to radically change the way we work. Instead, we chose to enhance the workflow we already use every day on GitLab—the coding platform we rely on to track changes to our projects’ source code.

The workflow is structured in this way, adapting a process that’s familiar to anyone who’s ever written code:

• One of our senior developers opens an "issue" on GitLab—essentially, they write a clear and detailed brief.
• Our AI Agent reads the request and writes the code needed to complete the task.
• The senior developer reviews the code, leaving comments and precise instructions—just like a standard code review.
• The AI Agent learns from the feedback and updates the code until it meets the requirements.
• Only then does the human give the final approval and merge the change into the main source code.

A oook to the future: AI, people, and clients

In our Customer Digital Solutions BU, we keep a constant eye on the evolving landscape of generative AI technologies for coding—with one clear goal: to integrate these technologies into our processes in an efficient and modern way.

But technology alone isn’t enough. Real transformation also depends on people. That’s why we’re already building upskilling programs for our developers, focused on AI coding—so they’re increasingly prepared to guide ever more capable AI models. We want our talent to stay the most competitive in the market—not despite AI, but thanks to it.

The underlying idea remains the same: let machines take care of the most repetitive tasks to free up time for human intelligence, creativity, and empathy. We want our teams to focus on what AI still can’t do—engage with clients on a highly technical level to guide them through a transformation process, understand their challenges in complex, chaotic scenarios, implement sophisticated software architectures, and build strong relationships that can only come from human interaction.

(Article by Fabio Piro, Head of Software Engineering & PMO)