Agentic Engineering: How We Are Using AI to Empower Development at Grupo Esfera

Writing less code is not the same as solving a problem better. Here is what we learned when we integrated AI agents directly into our development process.

What is agentic engineering, and why did we stop talking solely about "generating code"?

For a while, the conversation around AI and software development revolved almost exclusively around speed: AI writes code faster than a human team. This is true, and we see it every day in our active projects. Code that used to take days to produce now takes hours.

However, that speed introduces a new challenge: if a single developer has to manually review everything the AI generates, quality control becomes a bottleneck. The solution isn't to review faster. It is to shift the focus of software engineering.

At Grupo Esfera, we think of it this way: instead of having a person monitor all the code the AI produces, we define an agentic harness—a set of specialized agents, each with a distinct objective, that handle code generation, quality validation, and the execution of deterministic, mandatory tests. Software engineering shifted its focus: it is no longer about code generation, but about designing that harness—its stages, roles, and rules—to precisely direct the AI toward the product the business requires.

Put simply: core accountability now lies in the precise definition of requirements. That remains entirely a human activity.

So, is the developer disappearing?

No. They are moving to where it matters most.

The AI must be given appropriate, sufficient context at every step for the build to be cost-effective, along with precise prompts that guarantee code quality and test coverage. These three elements are still managed by the team. Before the AI writes a single line, a clear requirement must exist. We use collaborative practices to define scenarios and turn them into tests that drive development—this is not a technical detail; it is what keeps AI-generated outputs aligned with real business needs and our quality standards.

Speed is a consequence. Control is a decision.

How do we deal with the volume of code generated by AI?

It is a legitimate question: if code is produced much faster, where does validation fit in? At Grupo Esfera, we apply concrete technical practices to solve this: short iterations, test harnesses, and test-first approaches, among others. These strategies allow us to manage the complexity that still exists in the development process—because that complexity didn't disappear; it shifted. The team’s energy is no longer spent on coding, but on definition and validation.

The less human interaction a process stage requires, the more agentic it is. That is why human quality control is replaced or augmented by specialized agents taking on that responsibility—testing, static code analysis—alongside deterministic checks that do not depend on human interpretation.

A real-world case: Our modernization agentic harness

One of the questions we hear most often is how to migrate from a legacy architecture to a modern one without halting business operations. Migrating a legacy system is risky not because it is technically impossible, but because doing it poorly can paralyze operations while it happens.

Our answer is our proprietary modernization agentic harness. It is not just paper methodology—it is a set of rules, agents, tools, and workflows engineered to migrate an application incrementally, with minimal human intervention, and without interrupting live operations. The core principle is coexistence: the new application and the legacy system run in parallel, with the new system gradually absorbing functionality until the old one is completely replaced. There is no abrupt cutover.

And to be completely clear: this harness is not an empty promise. We are using it in a live, real-world case, migrating one of our own internal systems. We prefer to test our approach on ourselves before offering it to a client.

What stays the same: The Esfera Way

All of this rests on the foundation we built long before AI generated a single line of code for us: El Camino Esfera (The Esfera Way), our proprietary framework that integrates organizational agility with high-rigor software engineering. It was built from real-world practice, not theory—and that now includes the practice of working alongside AI agents.

Our core pillars remain unchanged: quality in decision-making, frequent delivery of visible results, and real impact on product performance. Automated testing, continuous delivery, and automated deployment are all still central. What changed is that part of this work is now carried out by specialized agents operating within a harness we design and oversee.

If your team is trying to figure out how to integrate AI agents into development without losing control over what reaches production, let's talk.

Do you want your team to apply these practices day-to-day?

We designed a course specifically for this: Evolution of Development Practices with Agentic Engineering

Let's discuss your project → Contact Us