AI has reached a point where it can write code faster than enterprises can govern it. However, while AI coding tools are making software implementation much faster, that doesn’t always mean that software companies are delivering finished products any faster. Major Australian software vendor Atlassian highlighted that divide on September 10 with its launch of new Jira and DX capabilities designed to improve the coordination and governance of AI agents across the broader software development lifecycle (SDLC).
The company’s 2026 AI SDLC study found that 94% of engineering leaders say their organizations use AI, yet only 6% say they have the systems needed to scale it across the full lifecycle. The problem is that writing code represents just one part of the software delivery process. Requirements still have to be defined, architectural constraints understood, dependencies managed, and changes tested before anything reaches production. Atlassian reports that, while 74% of engineering leaders say AI has accelerated coding itself, 78% say their teams still rely on traditional peer-review processes to handle the resulting code. In other words, most current implementations of AI have simply shifted the hard work further downstream.
Atlassian’s solution begins with context. Code Context, an AI context engine built on the company’s Teamwork Graph, is designed to give coding agents access to relevant information across repositories, work items and organizational knowledge. Meanwhile, Agent Context Controls let platform teams restrict which agents can operate in a given space and what they can see, thereby reducing the potential blast radius if an agent makes the wrong decision. Atlassian also cites a DX analysis finding that teams whose AI tools use the most Teamwork Graph context shipped roughly 64% more per developer.
The company is also moving beyond one-off prompts toward what it describes as a “governed agent loop,” where Jira can identify clearly defined unassigned work, delegate it accordingly to a coding agent for implementation and testing, and return a ready-to-review pull request. Moreover, its New Standards capabilities can apply organizational coding rules across repositories, while AI Review checks all changes against those standards before a human receives them. As such, software teams can automate more of the work leading up to a merge decision, while still maintaining ultimate control over it.
Of course, governance also depends on being able to reconstruct what happened later. Atlassian’s own research found that 88% of engineering leaders believe they need a governed system of work for AI, yet only 19% say they have built one. The clear majority of leaders and engineers struggle to reconstruct the reasoning behind an AI-assisted decision. That becomes a serious problem when investigating incidents, reviewing security issues, or simply explaining why a particular change was made.
For DevOps leaders, the main implication is that AI isn’t eliminating software-delivery bottlenecks, but simply moving them downstream. As AI code generation gets cheaper and faster, context, verification, traceability and measurement only become more important. The organizations that benefit most from coding agents will ultimately be those that can govern the work around the code, instead of merely generating more of it.