The author implemented a server-based system that integrates AI agents into Jira as virtual team members. These agents, assigned roles like frontend developer, backend developer, or analyst, handle specific tasks such as adding buttons, updating forms, or drafting requirements. The system uses a harness environment that manages project context, instructions, and access rights.
When a task is assigned to an agent in Jira, the server automatically processes it by loading project documentation, repository snapshots, and instructions. For coding tasks, agents generate merge requests in GitLab, including tests and documentation. For analytical tasks, agents create draft user stories in Confluence. Human oversight remains mandatory for code reviews, final approvals, and merging changes.
The system includes a backend office that tracks task states, artifacts, and token usage. It supports multi-task queuing, allowing users to assign several tasks without waiting for immediate completion. The infrastructure relies on PostgreSQL for state management and Redis for task distribution, enabling users to monitor progress and interact with agents asynchronously.
Agents can also interact via a chat interface to discuss ideas or analyze code without creating formal Jira tickets. The system requires registering agent accounts in Jira, Confluence, and GitLab, using personal access tokens for API authentication. While the backend tracks token consumption for cost estimation, the author notes that total operational costs also include infrastructure and human review time.
