Built to Deliver - Responsible AI in Practice

^ LISTEN TO THE INTERVIEW ABOVE ^ 

“That mix of experience, focus, and trust allowed the team to move quickly and to become the first team in the BIO initiative to pass both architecture and staging reviews, and ship both forms shortly after the government reopened.”

— Berni Bernardini Xiong

The Challenge

Four government contractors. A two-month deadline. Two forms delivered on time.

The project is called BIO, Benefits Intake Optimization, a Department of Veterans Affairs (VA) initiative to turn paper forms into digital ones so Veterans can get their benefits faster.

Eleven forms were split across the contractors. Agile Six had two to complete.

Digitizing a single form usually takes at least six months. This six-person team completed two in two.

Today, I’m talking with Berni Bernardini Xiong and Aaron Ponce about trust, a lean team, and what responsible AI in practice looks like.

On the Team: Meet Berni and Aaron

Berni: I’m a Team Coach at Agile Six, and I also served as a Program Manager for one of the four teams tasked to digitize key VA benefits forms as part of the Benefits Intake Optimization initiative, also known as BIO. My role sat at the intersection of delivery, coordination, and team health.

I helped keep engineering, design, product, and our VA partners aligned so that the team could focus on turning paper forms into online experiences that are easier for Veterans to complete and less manual for VA staff to process. Ultimately, we wanted to help Veterans get their benefits faster.

Aaron: I’m an AI Solutions Engineer on the BIO team, and I’ve been working with the Agile Six Innovation Lab on various projects. The intention of the lab is to explore AI and all the different tools available to us, from things that apply directly to the VA and to Agile Six, to things that don’t. It really is about getting familiar with the landscape and what’s out there, and then seeing how it all ties into the work we do.

Agile Six is doing a good job allocating resources to the lab, allowing us to experiment and become technically familiar with all the different aspects these tools offer. That’s given us a huge edge, and it’s already been implemented in VA work, so it’s proving successful.

We want to make sure that not only do we have the ability to leverage the technology, but we know when to use it and how to use it.

What the Team Was Up Against

Berni, you had less than two months to do work that usually takes six. Walk me through what you and the team were up against.

Berni: The constraints were intense. We had about two months to design, build, and deploy two production-ready digital forms. That kind of work normally takes about six months per form at the VA. Not only did we have that compressed timeline, but we also experienced a multitude of unforeseen circumstances: the government shutdown, multiple code freezes, and two cyber incidents that disrupted our access and pushed out our timelines.

Because this was VA.gov, the bar didn’t lower. We still had to meet full security, accessibility, and architecture standards and pass the same reviews as any other federal system. What made delivery possible wasn’t cutting corners. It was about being deliberate.

Did you ever think this was impossible?

Berni: I wouldn’t say impossible, but I might use the word improbable. Knowing what we know about how long it takes to digitize a VA form, given past experience, given what we see teams do, and given the size of the teams that also digitize forms, it seemed like a very improbable task to complete.

But I was also excited to see if it was something we could achieve.

What Made Delivery Possible

Aaron, with a team of six and everything you were up against, what do you think made delivery possible?

Aaron: We delivered because of the intentional staffing. It was calculated, making sure we had somebody with existing VA experience. There’s a lot to delivering a product on the VA that is non-code related. It’s documentation, it’s process, it’s organizing meetings. It’s not just the engineering work; there’s the whole design aspect.

We needed a group of folks who were very well versed in that process, so everyone could take on a role and keep momentum without slowing down.

It was the team structure that pretty much locked down the success. Things were getting done faster than we were writing them down, and having people who knew what needed to be done is how that was even an option for us.

Keeping the Team Moving

With that kind of timeline, how do you keep a team moving without getting buried in the process?

Berni: Because we were on a compressed timeline, we wanted to make sure we were making smart trade-offs without relying on any rigidity in a particular framework or heavy rituals. We still used structure. Just enough of it. We kept daily check-ins to stay aligned and communicate with each other regularly.

We also held weekly planning sessions, which helped us manage dependencies and priorities, as changes can occur frequently. We worked very closely with all of our stakeholders, which allowed us to adapt quickly when conditions changed. We had an agile, lightweight structure because it supported the work, but we wanted to make sure it wasn’t driving the work.

Agile Six encourages us to lead with trust and outcomes, not process for process’s sake. In this case, it enabled faster delivery without compromising quality or the Veteran experience.

What Stands Out Looking Back

When you look back, what stands out most for you?

Berni: How the team showed up for each other. From day one, there was real trust. There was psychological safety. People were honest about their capacity. If someone hit a blocker, they asked for help. If someone wasn’t able to pick up work, another person stepped in, even if it was outside their usual role.

As I think about this as a coach and delivery lead here at Agile Six, when I watched the balance of people moving fast while also still taking care of each other, that really pulled on my heartstrings. It was an incredibly wonderful thing to witness.

That doesn’t happen by accident. That comes from shared ownership, intentional team formation, and having aligned values. That’s the kind of work environment I’m so proud to be a part of. 

The Role of AI

Trust and teamwork drove this project, but how much of this came down to AI?

Aaron: Maybe 10 to 15 percent of what actually had to be done to get this project done. It was small in relation to the effort the group had to put in to make sure we were able to get through the staging review.

Let’s talk about the tool you were building in the Innovation Lab.

Aaron: I refactored the entire tool specifically for this project and turned it into a simple, multi-step tool that was focused on doing exactly what it was intended to do: scaffold a VA form from a PDF. What would have taken two weeks of work, the tool could scaffold a form in 30 minutes.

AI gave you the architecture, the bones, but where did it fall short?

AI was super useful when it came to the architecture and the initial implementation, getting things in motion. But it really came back to the team's cooperation and understanding of what the expectations were for that project.

I noticed the scaffolded website the AI would put out was full of issues that were not empathetic. It was using language that was verbatim from the PDF, and it takes one of us to go through that content and identify those issues. Even with all the context and all the guardrails, it was still using the wrong components and language that didn't make sense.

It just set us up with a starting point. That's how I see the whole thing. It moved us into a place where the group could actually get in, review the work, and collaborate in meetings. That’s where ideas emerged to make sure it flowed for the Veteran. That part was purely non-AI. That was conversational and team-related. 

The whole point of the tool was this: it’s not AI doing all the work. It’s a tool that the team uses to be more efficient. Knowing how to use the tool and when to use the tool is the key to success.

Berni: We didn't want to treat it like a shortcut. We always made sure we paired AI with the deep knowledge of VA systems, the forms infrastructure, and the VA design system. Team members weren't asking each other, How fast can we go with this tool? It was, How do we use this tool responsibly and actually make it work?

Responsible AI in Practice

So what’s that look like in practice?

Aaron: Responsible AI means knowing how much to allocate to the agent versus how much you should be involved, and knowing when you should be involved. We knew we needed to create validators and documentation for the validators to work with, and that took a lot of manual research to identify the rules we needed the agents to follow.

Then it was about knowing when to take over. We took over at the point where the infrastructure was laid out and the initial pages all existed. From that point forward, we were content focused, not just letting the AI run through and complete the entire thing.

One thing we’ve been working on in the Innovation Lab is governance and putting together a framework for Agile Six as a company to reference when using AI in their projects. Respecting safety, security, privacy, transparency. Is it reliable? I think those exercises actually contributed to the success of making sure we weren’t trying to do too much with the tools, and that we stayed as involved as we should be to make sure a good-quality app was being built.

Berni: Every single person on this team leveraged AI to optimize their work and streamline processes, not replace what they do.

The VA Partnership

The team was ready. The tool helped. But none of this works without the VA doing things differently, too.

Aaron: That was probably the number two most critical reason for the success of this project. The partners didn’t have us follow the standard process. They told us early on, if something is slowing us down or blocking us, talk to them and they’ll get it fixed.

That was the biggest hesitancy about the project. There’s a lot of process involved. Those processes are good to an extent, but they also create a lot of hurdles when you’re trying to move quickly. This partnership allowed us to bypass some of the processes that slow people down, and it worked because the team was experienced within the VA ecosystem.

That was a very specific relationship where they trusted us enough to make good decisions. Early on, step one was reviewing the forms and immediately going to them with trade-offs. This is feasible. This is not. We were able to work with them and say, what’s the MVP version of this form that’s acceptable for Veterans Day? They let us scale back some of the deliverables and create an enhancement period after the form was done. And they were able to provide air cover when we needed to move things faster.

Berni, how did you keep those relationships strong?

Berni: Everybody involved was doing their best inside a complex system that is the VA, and everybody’s under a lot of pressure. Being aware of that, being aware that our VA partners are also balancing speed with security, compliance, and long-term sustainability, that helped us remain collaborative partners instead of letting our frustrations get the best of us and turn things sideways.

We focused on what we could control. When we came into our collaboration touchpoints, we were prepared. We were clear about our assumptions. If we had questions, we asked them. We were transparent about risks and trade-offs. When things changed, our team adjusted quickly because we wanted to stay focused on moving forward.

The speed of our delivery ultimately came from the trust we built with those relationships. And that trust came from good intent, being able to adapt, and following through. 

Compressed Timeline: What Made It Possible 

What enabled this team to deliver within a compressed timeline?

Berni: Trust was the real accelerator here. Trust within the team. Trust with VA partners. And trust in our understanding of the VA system and its constraints. Because we understood those constraints, we didn’t try to power through them. We designed around them.

Domain expertise mattered. Discipline mattered. Ruthless prioritization mattered. AI helped us understand systems faster, but it was human judgment that made delivery successful for this team.

That mix of experience, focus, and trust allowed the team to move quickly and to become the first team in the BIO initiative to pass both architecture and staging reviews and ship both forms shortly after the government reopened.

When you combine these things, trust, deep expertise, focused delivery, and the responsible use of AI, even hard government problems can move faster. And in this case, it means Veterans don’t have to wait as long for the benefits they’ve earned. 

Looking Ahead

Aaron, what do you hope this proves for the future?

Aaron: Agile Six has been experimenting with leaner teams. When you have a smaller team with tools you can leverage to make that team productive in ways that would’ve originally required a larger team, that proved to be very successful in our situation.

I think the VA’s trust that each agency was going to be responsible and follow the guidelines they have in place, for code quality, governance, content, and making sure you follow the correct process to get your products delivered, I’d like to see more of that.

I’d like to see the VA embracing lean teams with more focused work, maybe some tighter timelines. Because in doing that, it keeps the partner very involved in making sure the deliverable is hit, and it doesn’t turn into a multi-year, bloated project.

Closing

We’ve proved that we can do forms in a month. I’ve never been able to work on a project in the past that was under nine months doing a form comparable in size. This proves that flexibility in how the product approach is done, not following that standard process I think most project teams are using, can really benefit timelines and deliverables. And you can do it with smaller teams now.

A lean, six-person team. Two forms. A two-month deadline. A partnership built on trust, and AI used to accelerate human judgment, not replace it. That’s what responsible AI in practice looks like, in a team that’s built to deliver.

If this story piqued your interest in a career with Agile Six, explore our open roles.