Integrating artificial intelligence into daily enterprise operations often proves far simpler than converting those financial inputs into verifiable, quantifiable returns.

Data reveals that the vast majority of executives surveyed by Atlassian find it challenging to pinpoint a definitive return on investment regarding their organization-wide AI deployments.

To bridge this persistent performance gap, Atlassian recently unveiled its Forward Deployed Engineering initiative. This project places applied AI professionals directly within client teams to architect and deploy customized AI frameworks aligned with unique corporate objectives.

Avani Solanki Prabhakar, Atlassian's chief people and AI enablement officer, sat down for an exclusive discussion to dissect why traditional implementation strategies frequently underperform and what is required to convert conceptual testing into concrete commercial value.

Interview responses have undergone light editing to optimize clarity and length.

Why is now the right time to launch Atlassian's new Forward Deployed Engineering (FDE) program?

There’s a massive gap between AI hype and actual business impact today. In our State of Teams report, we found that 85% of workers are using AI, but only 6% of execs can point to real ROI. That’s a stark reality.

Our approach to the Atlassian FDE program is focused on closing that gap by putting builders — true applied AI engineers, not consultants — directly into the customer’s environment. Beyond just handing over a tool or a model, we’re embedding the engineering expertise needed to turn complex operational needs into working systems.

It’s about moving from experimentation to production value, and we’re all in.

“There’s a massive gap between AI hype and actual business impact today.” Avani Solanki Prabhakar, Chief People and AI Enablement Officer, Atlassian

What does embedding engineers directly within customer teams enable that a more traditional AI implementation approach doesn't?

Traditional implementation often treats AI as a layer you just 'add' on top of old habits. But we’ve seen time and again that that doesn't work. By embedding engineers full-time, we get to the 'how' of the work.

By definition, Atlassian FDEs build inside of the customer’s own tooling and stack. We’re meeting them where they are. Better yet, the relationship is mutually beneficial. Our FDEs have access to our codebase, so if they notice a product gap while working with a customer, they can fix our AI platform, Rovo, directly.

That creates an immediate feedback loop, which helps both our team and our wider customer base. A traditional consultant might give you a roadmap, while an FDE gives you a production-ready solution that actually understands your unique business context. Context is the real accelerator here: When we ground AI in a company’s specific context, the quality of output jumps by 44% .

How do Forward Deployed Engineers determine which problems are worth solving with AI and which ones aren't?

We start with the desired outcome and work backwards from the customer’s problem, not forward from our product. Our FDEs look for high-volume pain points, like IT help desk routing, HR support or knowledge discovery where the coordination load is heavy.

These teams need to be ready to engage and commit to a partnership and identify the highest impact problems together, all with a shared goal of driving AI transformation.

This happens within the customer's existing technology stack, using a model-agnostic approach and the organization's own context. One thing I always come back to: If their workflow doesn't actually improve, we’re solving the wrong problem.

What have early customer engagements revealed about why some AI use cases reach production and deliver measurable ROI while others stall?

Most companies are in a 'throw darts and see what sticks' phase right now. It’s a necessary rite of passage, but use cases stall because enterprises are bolting on AI.

You can’t just drop AI into a complex, unique environment and expect to see ROI. The engagements that reach production start with a specific business goal. With one of our customers, a major technology company, that goal was developer experience. Our FDEs worked alongside their engineers to build agents that route engineers to the right answers and tools, and closed a gap in one of our enterprise software connectors along the way, a fix every customer now benefits from.

Moving to real impact also requires a builder mindset, comfort with ambiguity and a culture willing to be wrong and pivot quickly. In my capacity as chief people and AI enablement officer at Atlassian, I’ve seen time and again how the AI transition is as much about culture as it is tech.

“You can’t just drop AI into a complex, unique environment and expect to see ROI.” Avani Solanki Prabhakar, Chief People and AI Enablement Officer, Atlassian

C-suite leaders often stall AI adoption because of concerns around security and data access. How does the FDE program address the "black box" problem of AI governance?

Traditional governance wasn't designed for nondeterministic software. In the old world, you knew exactly what a system would do with a specific input. AI changes that, and it leads to the No. 1 objection we hear from the C-suite: What can this agent actually see, and who is in control?

We’ve seen technically brilliant pilots stall because these questions were treated as an afterthought. With FDEs, our approach is to treat governance as part of the architecture from day one.

We use a Read → Draft → Act progression. We start with 'Read' (search and summarization), move to 'Draft' (AI recommends, human approves), and only move to 'Act' once the organization is comfortable with the controls.

Underneath it all, the Teamwork Graph acts as a permission-aware foundation: It validates access in real-time so the admin always retains control. Whether it’s a global bank scaling users, or a healthcare provider needing to sanitize PII before it ever hits the model, we build the guardrails into the solution so security is the accelerator, not the handbrake.

The FDE program pairs Atlassian engineers with customers' own engineers. How do you make sure your customers become better equipped to build with AI on their own?

This is a collaborative model, not a dependency model. We require customers to contribute their own engineers to work alongside ours. By pairing up, the customer’s team learns exactly how to flex this muscle long after the engagement ends.

We prepare them for change management and provide training and playbooks so that they can walk away with repeatable patterns to scale their own programs. We also aim to provide them with a more connected graph, resulting in greater organizational context that they can build upon independently.

Success for us isn't a permanent seat at their table; it's seeing their internal teams move to that 'strategic AI collaborator' stage where they are shipping their own custom agents with little to no Atlassian oversight.

How does working directly inside customers' AI challenges inform Atlassian's broader products and platform?

It creates a product flywheel. Because our FDEs are true applied AI engineers, the code contributions, product fixes and reusable connectors they build in the field flow back into our core platform.

We’re learning about product gaps in real-time. If a regulated enterprise, for example, needs specific governance for Rovo, we build that, and then every other customer benefits from that improvement. We’re discovering the future of our own products by redesigning the actual flows of work alongside our customers.

“We’re betting big on the idea that the 'how' of work is the most important variable in the AI era.” Avani Solanki Prabhakar, Chief People and AI Enablement Officer, Atlassian

As the program scales, what will success look like for Atlassian and its customers?

We’re primarily focused on measuring the customer’s outcome and business value. The reality is most organizations are facing the same problems we have already solved, so we can share solutions that they can apply to their environments.

Also top of mind for me is the shift in proficiency: changing users’ perception of AI as a simple tool to a strategic partner. Broadly, I want to learn how we can scale 'intentional togetherness' between humans and AI agents.

We’re still in the early stages, and we certainly haven't nailed everything yet. But we’re betting big on the idea that the 'how' of work is the most important variable in the AI era.