Jira agents, the AI agents your team assigns work to, now show their work as agent sessions, so you can see what every agent is doing, where it is stuck and what is ready for review, all inside Jira. If your team assigns work items to Rovo, Claude, Cursor or GitHub Copilot, those Jira agents are already creating sessions. The question is whether anyone is watching them.
This guide shows team leads and Jira admins how to start, track, unblock and review agent sessions in Jira Cloud today. It also explains why some sessions never appear in your reports, and what Atlassian added at Team '26 Europe in October 2026. It is based on Atlassian's documentation and reviewed by Empyra's certified Atlassian consultants.
A Jira agent session is the record of one AI agent working on one Jira work item. It starts when someone hands the agent a task and ends when the agent finishes or is stopped. While it runs, the session shows one of three states: needs input, working or finished.
Before agent sessions, agent work was hard to follow. A developer could run an agent in a terminal or chat window, and the rest of the team saw nothing until a pull request appeared. Atlassian's answer is to make Jira the place where agent work is tracked, next to human work. On stage at Team '26 Europe, Atlassian co-founder and CEO Mike Cannon-Brookes put it simply: for 20 years, Jira has shown the work a team planned, and now it shows the work that is actually happening.
For the bigger picture of how Atlassian treats agents as teammates, with identity and governance, read our guide to Atlassian AI agents and AMP. This post stays practical: how to track Jira agents and their sessions on your own Jira board.
Jira agents come from Atlassian and from third party vendors. Each one creates a session when it picks up a work item.
|
Agent |
What it does |
Links to the work item automatically? |
|
Rovo agents (Atlassian's and your own from Rovo Studio) |
Drafts, summaries, triage, planning |
Yes |
|
Jira Coding Agent |
Turns a scoped work item into a pull request |
Yes |
|
Claude, Cursor, GitHub Copilot, SmartBear (cloud coding agents) |
Write code and open pull requests |
Yes, once the app is installed |
|
Agents run from Jira automation |
Any agent invoked on a work item in a rule |
Yes |
|
Rovo Chat |
Answers and drafts in chat |
Only if a work item is attached as context |
|
Local agents in a terminal or IDE (Claude Code, Cursor, Codex) |
Coding on a developer's machine |
Not yet by default. Needs the Teamwork Graph CLI, and full tracking is in early access |
Source: [Atlassian, Use the Agents dashboard](https://support.atlassian.com/jira-software-cloud/docs/use-the-agents-dashboard/). Checked 17 November 2026.
Atlassian also lists Devin, Factory and Warp on the Atlassian Marketplace, with OpenAI Codex coming soon. If you want to build your own Jira agents, see our Rovo agents services.
There are six ways to hand work to Jira agents. Each one starts a new session.
Before any of this works, a site admin has to turn on Rovo. Each Rovo agent also needs the Work item surface turned on in its configuration (Atlassian).
An agent only knows what the work item tells it. It reads the summary and description, and may use comments and attachments. Before you assign one:
One detail surprises most teams: only the person who assigned the agent, or moved the work item that triggered it, can interact with that session. Everyone else with access can see that an agent was triggered. The output stays private to that person until they share it as a comment.
My agent sessions gives you one list of every agent session you started, grouped by what needs you first. Atlassian made it generally available in July 2026 (Atlassian).
To open it:
Each session shows the agent name, its state, the work item key and summary, and when it was last updated. The most recent sessions are at the top.
|
State |
What it means |
What to do |
|
Needs input |
The agent has paused and needs an answer or more information |
Reply first. A count next to the group tells you how many are waiting |
|
Working |
The agent is still running |
Leave it, or redirect it if the approach looks wrong |
|
Finished |
The work is done and ready for review |
Review the output before you share or merge it |
Source: [Atlassian, View and manage agent sessions in Jira](https://support.atlassian.com/jira-software-cloud/docs/view-and-manage-agent-sessions-in-jira/).
Use the filter at the top to show one state, or All sessions to see every session across your spaces. Select a session and the work item opens in a side panel, where you can review the agent's work, make changes and comment without leaving the page.
Agent output is private to you until you choose to share it:
This review step is your quality gate. Treat agent output like a first draft from a new teammate: check it against the work item before it goes to the team.
The Agents dashboard shows agent usage, active users, cycle time and resolved work, so you can compare agent assisted work with the rest of your space. It is available on any paid Jira plan with Rovo enabled (Atlassian).
The catch: a session only counts once it is linked to a work item. If your dashboard looks empty or too low, check these five things in order:
There is one more gap to know about. Developers often update Jira from Claude or Cursor through the Atlassian Rovo MCP server. Jira records those changes as the developer, not as an agent session. Until local session tracking ships, that work is invisible in agent reports.
There are two ways to connect Claude and Jira. Developers can use Claude with the Atlassian Rovo MCP server, which works under their own name. Or you can install Claude Agent for Jira, which runs as an agent and creates agent sessions. This section covers the second route, the Claude Jira integration built for agent work.
This Claude Jira integration (Claude Agent for Jira, in beta) connects Jira Cloud to Anthropic's Claude Managed Agents. Once it is set up, you assign a work item to Claude. It writes the code, pushes a branch and opens a draft pull request in GitHub. The session shows on the work item while it runs (Atlassian).
You need three things before you start:
The setup takes four steps: install the app from the Atlassian Marketplace, create the Claude API key, create the GitHub token, then paste both into the app's Configure screen.
Check two limits of the Claude Jira integration with your security team first. Jira agents are not supported on HIPAA and FedRAMP instances. And Claude Agent for Jira doesn't support Zero Data Retention, because Anthropic keeps session history so long tasks can run. Start with one test repository, then widen access.
Claude sessions also have their own cost. Anthropic bills each session to your API account: model tokens plus $0.08 per hour of active session runtime (Anthropic). Your Rovo credit limits don't cover this, so set spend limits in the Claude Console.
Cursor and GitHub Copilot follow a similar pattern: install the Marketplace app, connect the account, then assign or @mention the agent.
At Team '26 Europe on 7 October 2026, Atlassian showed where agent sessions go next (Atlassian). Not all of it is available yet.
|
Capability |
What it does |
Status (17 Nov 2026) |
|
My agent sessions on the For you page |
Your own sessions, grouped by state |
Generally available |
|
Agents dashboard |
Agent usage and delivery metrics per space |
Available on paid plans |
|
Claude Agent for Jira |
Assign coding work to Claude |
Beta |
|
Team wide Agent Sessions view |
Every agent session for the team on the board, with filters for needs input, ready for review and done |
Closed early access |
|
Local agent session tracking |
Sessions from IDEs and terminals link to work items automatically |
Coming soon, closed early access |
|
Drag a session onto the board |
Turns an untracked session into a filled in work item |
Closed early access |
|
Session record and memory |
Saves the goal, changes, pull request, decisions and conversation so the next run starts with context |
Closed early access |
|
OpenAI Codex in Jira |
Assign work items to Codex cloud agents |
Coming soon |
To ask for early access, use Atlassian's Jira AI development waitlist. Atlassian hasn't published a date for general availability, so don't plan a rollout around one.
All of this needs Jira Cloud. Agent sessions don't run on Data Center, so if you're still on Data Center, start with an Atlassian cloud migration. For every Team '26 Europe announcement in one place, see our Team '26 Europe hub.
Jira agents use Rovo credits every time they run. That includes assigning an agent, @mentioning one and triggering one from a workflow or board column. The number of credits depends on the agent's configuration and the thinking mode it uses (Atlassian).
Rovo credit allowances apply from 3 December 2026. Admins get alerts at 80% and 100% of the allowance, and can track usage under Atlassian Administration, then Usage, then Meters. Third party agents such as Claude also bill through their own vendor.
Board columns and automation rules can start many sessions at once, so they are where costs grow fastest. To plan your budget, check how many Rovo credits each user gets and see where Atlassian AI credits really go.
Agent sessions only help if someone acts on them. Empyra's consultants suggest a short routine for teams starting out:
Put the review rules into your Jira workflows, for example a status a person must approve before agent code is merged.
Empyra is an Atlassian Platinum Solution Partner with 80+ certified consultants and specializations in Software Development and Cloud Migration. We help teams bring Jira agents into real workflows without losing track of output, access or cost:
Book an Atlassian AI consultation or talk to an Empyra consultant.