AI Release Manager

An AI-powered release manager that takes your code from issue to production automatically.

Comment

Inspiration

What We Built

We built an AI Release Manager, a multi-agent system that automates the software development lifecycle inside GitLab. Instead of functioning as a chatbot, the system behaves like a team of AI agents, each responsible for a specific role:

Planner Agent: Breaks down issues into structured development tasks Builder Agent: Generates code, creates branches, and opens merge requests Security Agent: Reviews code, detects vulnerabilities, and applies fixes Deployment Agent: Triggers pipelines and manages deployment

These agents are orchestrated in a flow triggered by real GitLab events, such as issue creation or pipeline completion. The system moves a feature through stages:

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AI Release Manager

Issue --> Planning --> Implementation --> Security --> Deployment

How We Built It

The project combines GitLab Duo Agent Platform capabilities with a lightweight backend:

GitLab Agents & Flows Defined custom agents using YAML and orchestrated them into an event-driven workflow. Webhook-Based Event System GitLab events (like issue creation) trigger our system automatically. Node.js Backend Handles tool execution such as: Creating branches Committing code Opening merge requests LLM Integration Used AI models to: Generate structured plans Produce code and test cases Analyze security risks CI/CD Integration Leveraged GitLab pipelines to validate, scan, and deploy generated code.

What We Learned

This project fundamentally changed how we think about AI in software development:

AI is most powerful when it takes action, not just generates text Orchestrating multiple agents creates far more value than a single “smart” model Event-driven systems are key to integrating AI into real workflows Developer productivity improves when repetitive coordination tasks are automated

We also learned how to design systems where AI interacts with real tools—turning abstract intelligence into tangible outcomes.

Challenges We Faced

  1. Orchestrating Multiple Agents

Designing a system where agents pass context and outputs between each other required careful structuring of inputs and outputs.

  1. GitLab Integration Complexity

Interfacing with GitLab APIs (branches, merge requests, pipelines) introduced challenges around authentication, sequencing, and reliability.

  1. Making AI Deterministic Enough

AI outputs can be inconsistent. We had to refine prompts and enforce structured formats (e.g., JSON) to make the system reliable.

  1. End-to-End Automation

Ensuring the system worked seamlessly—from issue creation to deployment—required debugging across multiple layers (AI, backend, CI/CD).

  1. Time Constraints

Balancing ambition with execution was critical. We focused on building a working MVP first, then iteratively improving it.

What Makes This Project Special

Unlike traditional AI tools, this project doesn’t just assist developers ,it acts on their behalf. It transforms GitLab into a platform where AI agents collaborate as a digital DevOps team, reducing friction across the entire software lifecycle.

From idea to production without manual handoffs.

Built With

  • axios
  • dot-env
  • express.js
  • gitlab-ci/cd
  • gitlab-rest-api
  • gitlabduoagentplatform
  • node.js
  • openai-api
Share this project:

Updates