Apps Developed and Maintained by AI: Where Are We Heading?

The evolution of Artificial Intelligence in software development is progressing at a dizzying pace. We are no longer talking just about code autocompletion or support chatbots, but about true autonomous agents capable of understanding, modifying, and maintaining entire codebases.

In this article, I want to share a technical deep dive into the new architecture we have implemented at Luciosoft for managing “dev triage,” exploiting the potential of new AI agents.

The Triage Problem

#

Anyone managing applications in production knows that triage of reports (bugs, feature requests, technical debt) is a time-consuming activity. Often, for every open issue, it is necessary to:

1. Analyze the context.
2. Identify the files involved.
3. Evaluate the complexity.
4. Assign the task or solve it directly.

This process takes precious time away from developing new features.

Jules Orchestrator and the Dev Triage Architecture

#

To address this challenge, we developed an architecture based on Jules Orchestrator, integrating Google’s Jules AI agent into our DevOps workflow.

The heart of the system lies in the devtriage module, which you can explore in detail here: Jules Orchestrator - Dev Triage

The Architecture

#

Here is a diagram summarizing the workflow we implemented:

graph TD
    A[User Report / Issue] -->|Webhook| B(Ingestion)
    B -->|Trigger| C{Analysis Module}
    C -->|Scan Codebase| D[Context Extraction]
    D -->|Prompts + Context| E[Jules Agent]
    E -->|Reasoning| F[Proposed Solution]
    F -->|Code Generation| G[Implementation]
    G -->|Run Tests| H{Verification}
    H -->|Pass| I[Create Pull Request]
    H -->|Fail| E
    I -->|Review| J[Developer Review]
    J -->|Approve & Merge| K[CI/CD Pipeline]
    K -->|Automatic Deploy| L[Production]
    L -->|Feedback Loop| A

How It Works

#

The architecture is designed to be event-driven and acts directly on our application repositories. Here is the simplified logical flow:

1. Ingestion: When a new Issue is opened or a bug is reported, a webhook triggers the orchestration.
2. Analysis: The system, via the devtriage module, analyzes the content of the report and scans the codebase to gather the necessary context.
3. Agent Activation: This is where AI comes into play. The Jules agent receives the context and the objective. Unlike past systems, Jules has the ability to “reason” about dependencies and the impacts of changes.
4. Implementation: The agent proposes a concrete modification, generating code that respects project standards.
5. Verification: The system automatically runs tests to validate the solution proposed by the AI.
6. Pull Request & Deploy: If the tests pass, a Pull Request is automatically opened. Once reviewed and merged, the CI/CD pipeline kicks in, triggering automatic deployment to production, closing the circle with immediate feedback on the update.

Towards Self-Managed Applications?

#

This implementation does not aim to replace the developer, but to empower them. By delegating the triage phase and the resolution of standard bugs to AI, one can focus on more complex architectural problems and business logic.

Just like in this image: while AI works for us resolving bugs and maintaining the infrastructure, we can finally enjoy some well-deserved relaxation… or almost!

The most interesting aspect of this architecture is the AI’s ability to maintain the application over time. Library updates, minor refactoring, and security fixes can be managed proactively by the orchestrator, leading us towards a model of “Apps developed and maintained by AI.”

We are only at the beginning of this revolution, but the results obtained with Jules Orchestrator at Luciosoft show us that the path is marked. Intelligent automation is no longer just a possibility, but a concrete technical reality.