How My AI Coding Agents Closed 200+ Jira Tickets in 8 Months
Custom AI coding agents in a medical software codebase: 200+ Jira tickets closed, under 10% reopened, 370+ merged pull requests and 420+ developer hours saved.
I'm a Lead Software Engineer and Technical Product Owner at medavis, where we build software for radiology. About eight months ago I started building custom AI coding agents that work through our Jira backlog like a teammate: they take a ticket, write the code and open a pull request.
Here is where that stands today.
The numbers
- 200+ Jira tickets closed
- Under 10% of them reopened
- 370+ merged pull requests
- 500+ preview deployments
- 420+ developer hours saved
In June 2026 our Group CTO presented the agent at the AWS AI Health Leaders Forum in Zurich, in the session "Everyone Is a Builder", as an autonomous AI teammate that ships feature pull requests on its own.
How it works
The workflow is deliberately boring. It runs on the tools the team already uses:
- A ticket goes to the agent in Jira.
- The agent reads the codebase and implements the change.
- It opens a pull request, and every pull request gets a preview deployment. Reviewers click through the actual change instead of only reading a diff.
- A developer reviews the pull request and merges it.
The agent also gets work from the pipeline itself. Our CI/CD is self-healing: when a build breaks, it files a Jira ticket that the agent can pick up and fix.
Measure what comes back
Counting closed tickets is easy and says little on its own. A ticket that gets reopened two weeks later cost more than it saved.
So every ticket the agent touches goes into a ledger: handled, closed, reopened or still open. The reopen rate is the number I watch, and it sits under 10%.
Standards live in the repo
An agent is only as good as the context it gets. It follows the same standards as the team, written down in the repository as an AGENTS.md file and documented code patterns. Good onboarding docs for humans turned out to be good onboarding docs for agents too.
Beyond code
The agents are the visible part. The bigger change was AI adoption across the company, about 200 people: AI-written release notes and QA test plans, a self-healing CI/CD pipeline, and AI skills that non-developers such as our QA team use on their own.
If you want to try this
Give the agent real tickets and track the reopen rate from day one. Once that number holds, the question in the room changes from "is this a toy?" to "what else can it take on?"