How we are revolutionizing code generation with AI at Pomelo

At Pomelo, artificial intelligence is no longer an experiment or a future promise; it is an essential part of our daily operations: we develop our own technology that redefines how we generate code, integrate clients, and build products.

This change directly impacts the speed, accuracy, and consistency with which we operate and scale solutions throughout Latin America.

Synthetic code built with AI from our infrastructure

One of the most significant impacts of this transformation is operational consistency.

Previously, each team could define their own testing and logging criteria. Today, these processes are standardized and automated. Logs are generated in a unified format with complete traceability, reducing friction, accelerating error resolution, and improving the experience of our engineering team, providing more speed and allowing more focus on business rules where much more value is added to our client's product.

This model allowed us to significantly reduce the time-to-market. The use of copilots like Cursor already contributes a 25% saving in development effort, supplemented by an additional 48% thanks to our reviewer agents that generate code and documentation from internal infrastructure that automates documentation, tests, and control of logs.

Conversational architecture, synthetic code, and collaboration between agents

We are building an infrastructure where specialized agents collaborate with each other and optimize code generation autonomously.

Each agent accesses its context and the capabilities of its own domain via an MCP (Model Context Protocol), which allows it to consult documentation, invoke specific tools, or interact with internal APIs in a structured and contextualized manner.

To collaborate with other agents, it uses the A2A (Agent-to-Agent) protocol, publishing an Agent Card in a common registry. This card describes its capabilities, endpoints, and skills, allowing other agents to dynamically discover it and collaborate without rigid integrations.

Thus, an agent that detects an improvement opportunity can consult its own context, discover other relevant agents, compose responses, or even propose automated changes.

And we go one step further: self-generating functions. We incorporate a discovery agent that detects repetitive processes, suggests improvement opportunities, and generates code proposals, leveraging the available context in each service's MCPs.

This is synthetic code, driven by an intelligent conversational architecture. It's operational efficiency at its highest level. And it's 100% developed by Pomelo.

MCP: our platform for intelligent agents

One of the most significant advancements in our engineering strategy is the creation of MCP (Model Context Protocol). On this platform, we develop intelligent agents that assist our clients in critical processes such as technical integration, reconciliation, onboarding, and operational support.

We are already using these Product Agents in areas like CX and Integrations, where they respond in real time to specific situations, accelerate implementation times, and free up human team capacity to focus on higher-value tasks.

Customer experience enhanced by AI

We incorporate artificial intelligence not only in the interfaces but also at the core of our stack: Our agents and MCPs allow us to:

• launch features faster
• detect errors before they occur
• automate repetitive processes
• keep the team focused on what matters most

For example: when a developer uploads code, it is handled by an automated pipeline with agents that review different aspects of development, such as tests, logs, documentation, security, and even detect unconsidered cases and propose the necessary code to resolve them.

This allows us to maintain a level of precision and coverage that was previously challenging to sustain in traditional workflows. We are already seeing improvements of up to 60% in functional coverage in some teams.

A new way to build technology, powered by Pomelo

AI has arrived to enhance talent. It allows us to focus on strategy, automate the repetitive, and stay at the forefront of technology. What used to take weeks of development and coordination among multiple teams can now be resolved in days, with great operational efficiency and standardized processes.

Behind the best cards, there is a lot of invisible technology. And today, a large part of that technology is artificial intelligence built by us with a clear objective: to continue building the most modern, secure, and efficient financial infrastructure in Latin America.