Atlassian upgrades AI coding agents for always-on software development

Atlassian is pushing Jira toward a more ambitious kind of automation: AI coding agents that can keep working in the background instead of waiting for each prompt. The company says the next phase of software development is “always-on” agentic AI, and its new features are meant to let engineering teams run those agents at scale without losing control over what they do.

The problem Atlassian is trying to solve is pretty familiar. More teams are using agents, but they’re also asking them to work longer, across more parts of the development cycle, and with less hand-holding. That creates obvious trust issues. Agents need context, shared memory, validation and a way to understand project standards if they’re going to do useful work instead of producing confident nonsense.

At the center of the update is Code Context, which uses Atlassian’s Teamwork Graph to give coding agents secure intelligence across complex, multi-repository codebases. The company is pairing that with Agent Space Settings and Agent Context Controls, so teams can decide where agents operate, what they can see and what they can do. In plain terms: management gets to treat them a bit more like employees with permissions than like loose chat windows.

There’s also AutoDev, which scans backlogs and turns work into code merge requests inside Jira. A standards system then checks the code, and a separate agent reviews merge requests against those standards to flag problems. On the documentation side, DevDocs can automatically create or update technical docs in Confluence straight from code repositories, so the docs don’t lag behind the code.

But Atlassian is not pretending this is magic. Every agent run generates an audit log with diagnostics, and the system can measure impact across throughput, quality, adoption and cost. There’s also an AI agent usage dashboard that shows who is using the tools, how they’re being used and what the outcomes look like at a team level. The pitch is simple: move from foreground coding tools to background agents that quietly chew through engineering tedium, while still leaving a trail you can inspect.