The landscape of enterprise software is shifting beneath our feet as we move beyond simple automation into the era of agentic workflows. To help navigate this transition, we are joined by Vijay Raina, a seasoned specialist in enterprise SaaS technology and a prominent thought-leader in software design and architecture. With a career dedicated to optimizing how large-scale organizations deploy and manage software, Vijay offers a deep understanding of the structural changes required to make artificial intelligence more than just a novelty in the developer’s toolkit. His insights are particularly vital now, as the industry grapples with the complexities of integrating autonomous agents into the full software development lifecycle.
In this discussion, we explore the critical shift from isolated AI prompts to deeply integrated organizational context, the staggering gap between initial AI adoption and true enterprise-wide scaling, and the new governance frameworks necessary to maintain code quality. We also delve into the practicalities of autonomous “agent loops” that handle everything from backlog grooming to pull requests, and how data-driven insights are finally allowing engineering leaders to measure the real-world impact of their AI investments on developer throughput and cost.
Many engineering teams find that AI agents struggle when they are isolated from the broader organizational context. In your experience, why is shifting toward a shared context through tools like the Teamwork Graph so much more effective than relying on individual prompts?
The frustration for most developers right now isn’t that the AI is unintelligent, but that it acts like a brilliant intern who hasn’t been given the employee handbook. When an agent works from an isolated prompt, it lacks the “tribal knowledge” found in architecture docs, historical requirements, and existing project information. By utilizing a system like the Teamwork Graph to provide Code Context, agents can finally see across multi-repository codebases to understand why a certain decision was made three months ago. The data is clear on the impact of this visibility: teams that leverage this deep organizational context ship roughly 64% more per developer compared to those working with local code prompts alone. It transforms the agent from a simple code generator into a teammate that can actually investigate defects and trace root causes with precision.
A recent study this year highlights a massive gap in the industry, showing that while 94% of engineering leaders are using AI, only 6% have the systems to scale it across the full development lifecycle. What are the primary hurdles preventing these organizations from moving past the experimental phase?
The gap we are seeing in 2026 is largely a result of the “control vacuum” that occurs when moving from one-off chatbot sessions to persistent, autonomous workflows. Most organizations are terrified of “hallucinations” or agents making unapproved architectural changes, which is why that 94% adoption rate is mostly restricted to low-stakes, ad hoc tasks. To bridge this, companies need robust Agent Context Controls that allow platform teams to explicitly limit which Jira or Confluence spaces an agent can touch. Without a way to ensure outputs align with internal standards and approved requirements, leadership will continue to hesitate to give agents more than 6% of the workload. It’s about moving from a “black box” approach to a governed environment where the agent’s sandbox is clearly defined and safely monitored.
We are starting to see the rise of “agent loops” that can autonomously scan backlogs and execute tasks. How do you envision these recurring workflows changing the day-to-day experience for a human developer, and what does the hand-off look like?
The day-to-day experience is shifting from manual labor to high-level orchestration, where the heavy lifting of clearing unassigned backlog tasks is handled by the Jira Coding Agent. Imagine starting your morning not by staring at a pile of minor bug fixes, but by reviewing a series of pre-prepared pull requests that an agent has already tested and drafted implementation plans for. This loop—scanning the backlog, executing the code, and testing—removes the “cold start” problem for developers, allowing them to focus on the final review and approval. It creates a rhythm where the human is the essential gatekeeper, ensuring that while the agent does the grunt work, the final code merge still meets the creative and strategic needs of the project. You can feel the tension leave a room when a team realizes they no longer have to spend hours on routine maintenance that previously clogged their sprints.
With both humans and agents working in the same codebase, maintaining quality is a major concern. How do features like “Standards” and automated AI reviews create a safer environment for this kind of collaborative development?
Maintaining a coherent codebase becomes exponentially harder when you have a dozen different agents and dozens of developers all pushing code simultaneously. By introducing a centralized “Standards” feature, platform teams can define coding rules once and link them directly to repositories, creating a “single source of truth” for everyone involved. This is bolstered by a separate AI review agent that acts as an automated quality auditor, checking pull requests against those organizational standards before a human even sees them. It’s a sensory relief for a lead architect to know that the foundational rules are being enforced by a machine that never gets tired or misses a bracket. This structured path from task definition to code review ensures that even as the volume of code increases, the quality remains consistent and governed.
Measuring the ROI of AI has historically been difficult for engineering leaders. How does the ability to track model-to-task fit and AI-driven throughput change the conversation regarding AI spending and engineering outputs?
For the first time, we are moving away from “gut feelings” about AI productivity and toward a cold, hard analysis of model-to-task fit through tools like DX for Agentic Development. Engineering leaders can now use an Agent Usage Dashboard to see exactly which agents are appearing in workflows and how those sessions relate to specific Jira work items. This allows a manager to look at the AI spend and directly correlate it to changes in delivery performance, quality, and even the adoption rate across the team. When you can see the cost per pull request generated by an agent versus a human, the conversation about scaling AI becomes a data-driven business decision rather than a technological experiment. It provides a level of transparency that more than 350,000 customers, including 85% of the Fortune 500, are now demanding as they integrate these tools into their core operations.
What is your forecast for the evolution of agentic development over the next year?
I expect that by this time next year, the “agent-to-developer” ratio will become a standard KPI for every high-performing engineering organization. We will see a shift where the majority of software teams move away from manually assigning tasks to individuals and instead move toward “intent-based” planning where the system automatically delegates work to the most appropriate agent or human. As the bottleneck of organizational context is solved by tools like the Teamwork Graph, the “intelligence” of the model will matter less than the “integration” of the model into the existing system of record. We are heading toward a future where the software development lifecycle is a seamless, self-correcting loop, and the role of the developer will be redefined as a “System Architect” who manages a fleet of specialized agents to deliver value at a pace we once thought was impossible.
