AI · Case study
JiraFlow Agent
Agent that picks up a Jira ticket, makes the change in the repository and opens a pull request against it. One branch per ticket, with human review still the merge gate.
- Team
- Solo
- Timeline
- January to February 2026
- Stack
- Python · React · Vite · Jira API · GitHub API · LLM Agents
Situation
A large share of any backlog is small, unambiguous tickets: a copy fix, a colour change, a missing prop. Each one still costs a context switch: read the ticket, branch, edit, push, open a PR.
Task
Automate the mechanical path from ticket to pull request for changes small enough to describe completely, without letting an agent write to the main branch.
Action
I built an agent that reads a Jira/SCRUM ticket, locates the files it describes, applies the change, and opens a pull request from a dedicated `ai/scrum-<id>` branch whose commit message cites the ticket. The repository history is the proof, because every change arrived as a reviewed PR from its own ticket branch. The design point is the boundary: the agent proposes and a human merges, so a wrong edit costs a rejected PR rather than a broken main branch.
Result
Small tickets close as reviewable pull requests without a context switch, and every automated change is traceable to the issue that asked for it.
Ticket → PR
fully automated
1 branch
per ticket
Human
stays the merge gate