AI is accelerating implementation. The engineering teams who benefit most will connect context, orchestration, and accountability across the work around code.
AI can now help engineers turn ideas into working code in minutes, but software delivery has always been more than just producing code.
Engineers need to break down ideas into plans and work items, triage whether they’re ready to build, what constraints and dependencies apply, verify whether code changes do what they should, and still so much more. These tasks require context, guardrails, and a trusted system of record. Without one, AI simply helps individual engineers generate code, and not teams delivering features.
This is the gap at the center of The Agentic Pivot, Atlassian’s new research report on the next era of software delivery. Across more than 1,100 engineers and engineering leaders, 94% of engineering leaders say their organizations use AI in some capacity, but most of that usage is still supporting individual tasks such as coding, debugging, and documentation. This results in significantly increased code volume, but at the expense of adding pressure to the bottlenecks before and after code implementation.
Engineering leaders are aware of this. In our survey, 88% say they need a governed engineering system of work for AI, while only 19% say they have built one.
The next step is to formalize AI-assisted delivery: give engineers and agents a governed system of work that carries the context, evidence, and accountability they need across the lifecycle.
Code generation is accelerating, but trust is not keeping pace
Consider a common engineering flow: An issue is triaged, a developer or agent begins implementation, teammates review the pull request, the change is deployed and monitored. In an ideal world, each step carries the intent, relevant context, decision history, and verification evidence needed by the next.
In practice, that information is often scattered: Acceptance criteria lives in a ticket, standards in documentation, critical architecture decisions in a Slack thread. During review, engineers need to connect the change to its requirements, constraints, prior decisions, and verification evidence. When that context is not carried with the work, they have to reconstruct it themselves, making review harder to scale as implementation accelerates.
The research makes this tension visible: 74% of leaders say AI is accelerating code generation, but 78% say their teams still rely on the same traditional peer-review process for the velocity and volume of AI-generated code. Without formalized context, governance, and review practices that can keep pace, teams will struggle to turn faster code generation into higher confidence in what reaches production.
In our survey, only 15% of engineers and 25% of leaders are very confident they could reconstruct the reasoning behind an AI-assisted decision six months later.
Capturing what agents do and why
When we make code changes, we explain what’s changing, why, and most importantly, we write it down.
The same must become true for agents: We need a system of record allowing agents to write back and document what it did, the context it used, what decisions it made, what remains unresolved, and what should happen next.
Just as we document our work as engineers to create a record that others can reference, this agentic system of record bccomes essential when AI-assisted changes need review, an incident requires investigation, a customer escalation arrives, an auditor asks a question, or a new team member needs to understand a decision made months earlier. It makes agentic work understandable without making it less accountable to humans.
It’s also more efficient. In our internal testing, agents using Atlassian Teamwork Graph context—a data layer connecting an organization’s people, work, and knowledge—improved answer quality by 44% while using 48% fewer tokens.
Three moves for the next phase of AI software delivery
As AI compresses code implementation, the work before and after code becomes more important. For engineering leaders, the job now is to build the foundation that makes intent, ownership, context, and governance visible across software delivery in three ways:
- Make intent and context explicit before code is generated. As implementation becomes cheaper, building the wrong thing (due to ambiguous requirements and fragmented knowledge) becomes costly. Teams need specifications, architectural guardrails, engineering standards, and test expectations to be first-class inputs to development.
- Extend AI beyond code generation. Two of the biggest opportunities for AI are in strengthening testing and quality, and improving observability and incident triage. Teams already identify quality and testing as a top priority if they were given additional capacity, yet AI-automated guardrails remain a minority practice. And as teams ship more AI-assisted code, they need better ways to detect incident signals early, understand production behavior, and close the learning loop faster. The result is a system that helps teams ship with greater confidence.
- Make agent work traceable and observable. Teams need to see which agents are active, what context they used, where they are blocked, and how key decisions were made. Visibility often seems like overhead, but it’s necessary if teams are to trust the work delegated to agents. It’s also crucial if leaders are to understand and prove whether agentic work is delivering outcomes and measurable ROI.
For more insights on how engineering leaders see AI changing software delivery, including insights on the changing role of engineers, where to invest capacity gained from AI, and more, read The Agentic Pivot.
