Atlassian has expanded Jira so that engineering teams can plan tasks, assign them to programmers, track their execution, and measure costs all within a single environment. The platform supports Claude Code, Cursor, and GitHub Copilot, includes its own agent capable of generating change request submissions, and is preparing for direct integration with OpenAI Codex.
The key features of the new AI-powered Jira in 20 seconds
- Jira will be able to assign work to different agents and display which tasks are blocked or pending review.
- Teamwork Graph will provide context from Jira, Confluence, Slack, and repositories.
- The proprietary agent will operate in an isolated environment and generate pull requests.
- Atlassian will also measure tokens, expenses, and estimated costs per code change.
The company aims to address a contradiction emerging in development departments. Programmers increasingly use assistants and AI agents, yet this growth hasn’t translated into a proportional improvement in software delivery capabilities.
A longitudinal study by Atlassian and DX reports a 65% increase in AI usage among engineers, while the average improvement in development speed hovers around 10% and peaks at about 15% in the analyzed organizations. These figures are from research sponsored by Atlassian and do not represent a universal sector metric, but they help explain the focus of the new Jira.
The issue is no longer solely about writing code. Before starting, it’s necessary to understand what needs to be built, review previous decisions, identify dependencies, and turn a business request into technical requirements. Afterwards come testing, security review, documentation, approval, and deployment.
While agents can generate code quickly, they might also interpret tickets too literally, ignore architectural restrictions, or produce seemingly valid solutions that still require hours of review. Atlassian intends to use Jira as a coordination layer between people and these agents.
Jira stops being just a record of who does what
For over two decades, Jira has served as a bug tracker, story manager, project organizer, and workflow tool. The new approach broadens this role: an agent becomes a participant to whom work can be assigned, just like a team member, with actions linked back to the originating ticket.
Users will be able to assign tasks directly to Claude Code, Cursor, or GitHub Copilot. Integration for assigning work to Codex will come later, although Jira already allows opening specific tickets in the local Codex app with main information preloaded.
| Capability | What the team can do |
|---|---|
| Assignment to agents | Send a ticket to Claude Code, Cursor, or GitHub Copilot |
| Jira Coding Agent | Generate code within Jira and prepare a pull request |
| Centralized sessions | See which agents are working, blocked, or awaiting review |
| Automations | Send errors, tests, or documentation to background agents |
| Agent templates | Create projects with predefined statuses and workflows |
| Cost reporting | Relate token consumption and expenses to projects and changes |
Jira Coding Agent is Atlassian’s proprietary tool to execute this work. It can read tickets, access authorized info in Jira and Confluence, connect to the selected repository, and work within an isolated cloud session.
The agent can run Bash or PowerShell commands, modify code, and push changes to a branch. It can also create a draft pull request, but not merge it autonomously. Final review and approval remain with the team.
Permissions inherit from the user who starts the session. It shouldn’t access Jira, Confluence, or repositories to which that user doesn’t have access. Atlassian also states that each session clones only the selected repository within the isolated environment.
| Actions of Jira Coding Agent | Specified scope |
|---|---|
| Query Jira and Confluence | Within user’s permissions |
| Access code | Only the selected repository |
| Execute commands | Within the isolated cloud environment |
| Create a branch | Allowed if authorized by the user |
| Open a pull request | Can create a draft |
| Merge changes | Not permitted |
| Access local environment | No |
| Delete Jira or Confluence info | No |
This setup reduces some risks but doesn’t eliminate the need to review code. An agent might introduce vulnerabilities, modify more files than necessary, or generate self-validating tests that overlook critical cases.
Session visibility aims to prevent work from being scattered across terminals, private chats, and disconnected tools. Jira will display which agent received a task, what actions it took, and where human intervention is needed.
Teamwork Graph aims to solve the business context problem
The core of the proposal is Teamwork Graph, the layer where Atlassian links tasks, documents, people, objectives, code, and past decisions.
An agent might get a ticket saying “fix the payment process,” but that phrase doesn’t specify which services are involved, what the team decided six months ago, what compliance requirements exist, or which component should not be altered. This knowledge is often scattered across Jira, Confluence, Slack, GitHub, and internal conversations.
Teamwork Graph seeks to prepare a knowledge set that helps the agent understand the task before writing code. Internal tests from Atlassian show that enriched agents achieved 44% higher accuracy and used 48% fewer tokens than those without this layer. These results are based on Atlassian’s internal experiments and do not detail all models, repositories, or criteria used.
| Information sources | Potential contributions |
|---|---|
| Jira | Requirements, responsible parties, statuses, and priorities |
| Confluence | Specifications, decisions, and documentation |
| Slack | Conversation context where requests emerged |
| GitHub | Code, branches, changes, dependencies |
| Loom | Video, voice, clicks, and visual explanations of a task |
| Jira Product Discovery | Customer needs and product decisions |
Jira Planner will use this information to convert complex projects into structured technical specifications. It can review the codebase, Jira and Confluence histories, and team context before generating a document in Confluence that can be reviewed by a human or an agent.
This feature is not yet widely available. Atlassian has opened a waitlist for early access, and its effectiveness will need evaluation when deployed in organizations with incomplete documentation, legacy projects, and large repositories.
Jira for Slack will allow creating tickets and assigning tasks within a conversation by mentioning @Jira. The system can transfer relevant message content to the work item, synchronize new messages as comments, and prevent the agent from receiving a separate summary of the original discussion.
Loom will be used to turn screen recordings and spoken explanations into structured instructions. A responsible person can show an error, locate it, and describe the expected behavior. The platform will extract screens, clicks, links, and voice to prepare a plan that later can become Jira tasks.
The goal is not just to provide more data but to select only the relevant information. Sending complete documents to an AI model can increase costs, fill the context window, and make it hard to distinguish important decisions. The effectiveness of token reduction depends on Teamwork Graph’s ability to filter and relate information properly.
Automate errors, tests, and vulnerabilities without losing control
Jira will allow integrating agents into its automation rule builder. Teams can set rules so that certain simple errors, missing tests, or documentation tasks are automatically assigned to an agent.
These agents will work in the background and notify when a pull request is ready for review. Atlassian envisions this model to also help eliminate outdated function indicators, fix known vulnerabilities, or update documentation after changes.
The company claims to have reduced up to 80% of the time spent on specific repetitive tasks within some Jira teams. These are internal results tied to particular workflows and do not indicate a universal reduction across all development cycles.
| Work that can be delegated | Control to maintain |
|---|---|
| Small fixes | Change review and testing |
| Test generation | Coverage and quality verification |
| Known vulnerabilities | Security validation |
| Updating documentation | Accuracy review |
| Removing old code | Dependency confirmation |
| Repetitive changes | Limits on repositories and files |
While automation can increase the volume of prepared changes, it might also create bottlenecks in review. If multiple agents generate requests faster than developers can examine them, backlog accumulates and the risk of unreviewed changes being approved grows.
Atlassian aims to make this visible through a unified session view. Managers can check what’s running, which agent stopped, which change needs review, and how long it’s been waiting.
Additionally, Atlassian introduces in DX an report that links AI tool expenses—Claude, Cursor, GitHub Copilot, and others—to engineering outcomes. It estimates AI costs per pull request and aggregates token consumption across providers.
Measuring the cost per change can be useful but doesn’t fully indicate quality. A inexpensive pull request that introduces technical debt or requires extensive review may be worse than a more costly one that solves a complex problem effectively. Companies will need to balance spending with cycle times, issues, revertions, and deployment results.
What’s available now and what’s coming later
Atlassian states that the agents for Claude Code, Cursor, and GitHub Copilot, Jira for Slack, Jira Coding Agent, automation, templates, and session view are already available to paying Jira Cloud customers without needing a separate add-on.
Support documentation adds some caveats: Jira Coding Agent requires AI features to be enabled, Rovo credits, GitHub Cloud or Bitbucket Cloud connection, and a compatible edition. “No additional cost” doesn’t mean unlimited use or automatic availability in all configurations.
| Function | Status |
|---|---|
| Claude Code, Cursor, and GitHub Copilot in Jira | Available |
| Jira Coding Agent | Available in compatible plans |
| Jira for Slack | Available |
| Automation with agents | Available |
| Session view | Available |
| Jira Planner | Waiting list for early access |
| Rovo for Microsoft Teams | Early access |
| Direct assignment to Codex | Coming soon |
| DX AI Cost Management | Available for Atlassian DX customers |
The distinction between creating a ticket and assigning it as an agent should be clear. Jira can now launch Codex, Claude Code, Cursor, or GitHub Copilot locally with the ticket context loaded. The integration for Codex as an assignable agent directly from Jira is still pending.
Atlassian aims to turn Jira’s historical role as a work record into an advantage in managing agents. Its strategy isn’t just to have the best programming model but to control context, assignment, supervision, and measurement of all these elements.
Success depends on Teamwork Graph accurately representing each organization’s reality. When tickets are incomplete, documentation is outdated, and key decisions are only in private conversations, the agent inherits these issues.
Jira can provide a common interface for people and automated systems but doesn’t replace the need to define requirements, maintain reliable knowledge, and review what’s in production. Atlassian’s new approach focuses on ensuring these less-visible tasks aren’t left out of automation.
Frequently Asked Questions
What programming agents can Jira use?
Atlassian supports Claude Code, Cursor, and GitHub Copilot. Codex can be invoked from a ticket with preloaded context, but its direct assignment as an agent in Jira will be available later.
What is Jira Coding Agent?
It’s Atlassian’s proprietary agent for code generation from tickets. It runs in an isolated cloud environment, can create branches, and draft pull requests, but cannot merge changes autonomously.
What does Teamwork Graph add to agents?
It links tasks, documentation, conversations, code, people, and decisions to provide a broader business context. Internal tests show improved accuracy and reduced token consumption.
Are these features free?
They are included in compatible paid plans, but some require Rovo credits, AI activation, and repository connections. DX AI Cost Management is available to Atlassian DX clients.
