Starting a business is supposed to be about the idea.
The vision. The risk worth taking. Instead, for business owners in this state, it meant weeks of fragmented research across a dozen agency websites, phone calls to offices that may or may not have the answer, forms filled out wrong, fines for licenses nobody told them they needed, and a creeping suspicion that the system wasn't built for them. It wasn't. It had never been designed at all.
This project set out to change that, and it did: a greenfield, AI-powered platform, now in production, that consolidates every requirement, permit, regulation, and license across state agencies into a single, personalized experience for business owners. A one-stop portal built on a simple promise: we did the work so you don't have to.
That promise only works if people can trust what the AI tells them. So the hardest design problem wasn't the interface a business owner sees. It was deciding where human judgment had to sit between AI-generated content and a decision about someone's livelihood.
Building it meant designing three interconnected things at once. A content-driven public website. A wizard that asked business owners the right questions to generate a personalized compliance checklist. And an entirely new internal governance platform for state agency administrators, a role and a workflow that had never existed before.
As Design Principal, I led the full effort and managed a team of two. Jessie Betz, a contract designer, was my main design pair, focused on carrying the brand identity into the products themselves. I brought in Nahin Shah, a visual designer from Thoughtworks, to work with Jessie on the brand identity. Three streams running in parallel, converging into one product.
Starting with the landscape, not the screen
The project opened with volumes of agency documentation: regulatory language, existing site content, disconnected processes across a dozen departments. Rather than digest it manually, I had the team lean on AI tools early and aggressively to synthesize documentation, surface patterns, and get oriented faster than a manual read-through ever could. That synthesis gave us a foundation to design against from week one, instead of spending the first month making sense of what we'd inherited.
Alongside that, a dedicated research stream ran AI-accelerated studies on both sides of the platform, customer-facing and admin-facing, digging into the needs, pain points, and goals of people navigating business compliance and of the administrators who would eventually manage it. AI made the research move fast: consolidating findings, surfacing trends, extracting insight at a pace that would have taken weeks by hand. But speed wasn't the whole story. The tools could tell us what the data said. They couldn't sit across from a business owner and understand why it mattered, or carry that weight into a design decision. That tension, AI for velocity and human presence for meaning, became one of the clearest lessons of the project, and it shaped how I used AI everywhere else on the engagement.
A branding process, proven for the third time
Branding hadn't been scoped into the original project, and I flagged early that it needed real attention. I brought Nahin Shah on to work it with Jessie, running the same branding process I'd developed on the Lincoln Unlocked museum project and carried into the wholesale distributor app. This was the third time that process had been put into practice, and it held up: a repeatable way to take a team from an undefined brand, in this case a state agency's public-facing identity, to a resolved visual language.
Nahin and Jessie ran the branding track together under my guidance, exploring visual direction with Gemini and Midjourney: Gemini for prompt refinement and color exploration, Midjourney for generating and pressure-testing directions. We intentionally tested a range of AI tools along the way to see which ones actually earned a place in the workflow. The track ran in parallel with delivery rather than blocking it, and folded back into the core product once the visual language was resolved. That was a genuinely rewarding moment, when brand, research, and experience came together as one product.
The problem nobody had solved yet
The AI at the center of this platform pulls from the same agency sources business owners have always had to navigate themselves. It consolidates that content and presents it for human review and approval before it ever reaches a business owner. The administrators, representatives from each state agency, are the human in the loop: the trust layer between AI-generated content and a business owner making a real decision about their livelihood.
But these administrators had never done this before. There was no existing workflow, no existing role, no existing mental model to design toward. I built it with them, through research, co-design sessions, and continuous feedback loops. Every decision had to balance two competing needs: enough rigor to ensure accuracy, and enough simplicity that administrators could get in, get going, and get out. A quagmire of an admin experience would have undermined the entire trust architecture of the platform.
Accessible by requirement, and by conviction
As a public government platform, the experience had a hard WCAG 2.1 AA requirement across all three surfaces: the public site, the wizard, and the administrator tools. I treated that as a design input from the first wireframe rather than an audit at the end, and it shaped decisions well beyond the interface. During branding, it meant walking away from a set of identity directions built on gradients we liked, because they couldn't meet color contrast requirements across the product. That's the trade I'd make every time. For a platform whose whole promise is “we did the work so you don't have to,” an experience that excluded anyone would have broken that promise before it started.
Designing at the speed of AI, with the judgment of a human
While branding and research ran their tracks, the core delivery stream moved on skeletal UX: early structural design work that let the team keep pace while the visual language was still resolving. As product requirements came through prioritization, I set up a rhythm for turning them into design: UX Pilot for breadth, generating a wide scatter of directional variants fast, then Figma for depth, picking those explorations apart, applying craft, and bringing in the judgment no AI tool has. I think of that second phase as the humanification step, the place where a generated skeleton becomes something that respects the person using it. I also brought AI into how the team worked together. When a static screen couldn't answer the question, the design team built working prototypes in Figma Make while the development team built in Cursor. Design and dev would sit down over those Cursor prototypes to settle experience and interaction questions on something real, instead of debating pictures of it.
Early on, that filtering was directly informed by the research stream, with direction, feedback, and iteration in a tight loop. Once the research wing wound down, the filtering became more intuition-led. But that intuition was never starting from zero. The early research had built a foundation of understanding, what business owners and administrators actually needed and where the real friction lived, and every downstream decision, even the ones made on instinct months later, was built on that ground. Intuition isn't a substitute for understanding. It's what understanding turns into once it's internalized.
Leading across the full spectrum
I owned the design of everything on this project, the public website, the wizard, the checklist, and the administrator platform, while managing Jessie Betz and Nahin Shah across delivery and branding. I was also the primary design voice in a fixed-bid engagement, which meant communication with the Product Manager wasn't just important. It was the mechanism that kept the whole project healthy. Fixed-bid projects punish ambiguity, so I became deliberate, almost surgical, about how I presented design decisions: always anchored to intent, always connected to the problem we'd agreed to solve, always leaving the PM with clarity rather than questions.
At the same time I was operating at delivery altitude, I held the long-term strategic view. How does this platform scale as more agencies come on board? How does the administrator experience evolve as the AI gets better, and as it earns enough trust to need less review? How do we design a content architecture flexible enough to handle the full complexity of state government without collapsing under its own weight?
That's the work. All of it, at once, every day.
What this project taught me
There's a version of government digital experience that earns trust by being competent. Clear, accurate, fast. That's the floor. This platform reached for something higher: treating a business owner's time, confusion, and anxiety as design inputs as important as any technical requirement. It's now in production, with agency administrators using the governance tools to review what the AI produces. When a business owner gets their personalized checklist and feels confident rather than overwhelmed, that isn't a feature. That's the mission.
It's also where I sharpened something I carry into every AI-augmented project: AI is extraordinary at breadth, synthesis, and speed. It is not a substitute for sitting with a person and understanding what they're going through, and it isn't a substitute for the judgment that turns research into a decision. The best work happens in the handoff between the two, with AI clearing the ground fast enough that the humans on the team can spend their time where it matters. Designing that handoff, deciding what the system does on its own and what a person reviews first, is the design problem I find most worth solving right now.