Atlassian’s approach suggests that work-management systems may become a control plane for agentic development, supplying architecture intent, task boundaries, review rules and repository-level policy. That can improve consistency and reduce token-heavy codebase exploration, but it also creates a dependency on the quality of metadata in Jira, Confluence and connected repos. If stories, ownership models, architectural constraints and coding standards are incomplete or stale, agents will automate ambiguity rather than remove it.
The governance angle matters just as much. Context controls and pull-request review agents can help limit oversharing and standardize checks, but practitioners should test whether those controls remain enforceable across mixed environments where multiple AI tools, CI pipelines and repository platforms interact. The practical concern is less whether one agent is well governed than whether auditability, permissions and policy inheritance survive handoffs between vendors and stages of the pipeline.
Before scaling this model, teams should examine three things: whether their development metadata is structured enough to guide agents, whether repository and identity controls can express least-privilege access for machine actors, and whether review workflows can prove why an agent made a change. Without those foundations, “agentic engineering” risks becoming faster automation with weaker traceability.
Atlassian today extended the capabilities of its portfolio to provide teams of artificial intelligence (AI) coding agents with the level of context needed to build and deploy applications in production environments at much higher levels of scale.
Code Context, built on Atlassian’s Teamwork Graph, makes it possible for AI coding agents to generate more reliable output by, for example, vetting ideas for architectural feasibility before a line of code is generated, while Agent Context Controls makes it possible for DevSecOps teams to govern which agents can operate in a space and exactly what they’re allowed to see.
Additionally, Atlassian is adding to its Jira and Confluence platforms for managing software development projects an ability to continuously scan for well-defined, unassigned work items that can be delegated to a Jira Coding Agent that creates a pull request for review. An AI Review capability provides a dedicated agent on every pull request, checking it against those coding standards that DevSecOps teams can now map to repositories to automatically apply consistent quality guardrails across every agent and human developer.
Ming Wu, head of engineering for the DevAI organization at Atlassian, said that while AI coding tools have been widely adopted, most DevOps teams are not using them extensively across the software development lifecycle (SDLC). Achieving the next level of agentic engineering will require tools and platforms that surface the context that AI agents will require to work as a team, she added.
Right now, there are simply too many gray areas where the level of visibility being given to AI coding agents isn’t sufficient to complete tasks assigned, noted Wu.
Mitch Ashley, vice president and practice lead for software lifecycle engineering at the Futurum Group, said Atlassian is focusing on where agents get context and permissions, a different context than coding assistance. Agent deployment is limited by what teams can observe, control, and prove, and that ceiling sits in the systems where work is defined, he added. The open question is whether context and controls scoped to one vendor’s graph hold up when agents from three other vendors touch the same repository, noted Ashley.
It’s not clear how deeply AI coding tools and agents have been embedded across the SDLC, but it’s clear that software engineers, instead of focusing on a particular task that previously might have been assigned to them, will need to manage projects at a higher level. In effect, every developer is evolving into a manager of an application development project where tasks will be completed by one or more AI coding agents.
While AI coding agents continue to improve as the underlying models they rely on become more powerful, each DevSecOps team will need to become more adept at constructing workflows in a way that consumes tokens most efficiently. By relying more on platforms such as Jira to provide context, the number of tokens that would otherwise be consumed by an AI agent as it navigates a codebase can be significantly reduced. The challenge is striking the right balance between the capabilities of the AI model and the tools and platforms being accessed via the harness that has been applied to the AI agent.
In the meantime, however, DevSecOps teams would be well-advised to start re-engineering their workflows for the agentic AI era now rather than simply adding AI agents to existing workflows that are usually rife with bottlenecks.
Atlassian Aims to Fill Context and Governance Gap for AI Coding Agents
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.