In August 2025, a major reorg reshaped the Power Apps organization, and I stepped in as Design Lead for the newly formed Developer Agents team. Nothing existed on day one.
There was no inherited design, no established product surface, no settled team structure, and no shared definition of what "good" looked like. The business wanted to move decisively toward an AI-first way of building apps: a 0-to-1 product that would let the organization ship faster, experiment in production, and learn from real customers.
The mandate was bold: an enterprise-grade app-building experience where specialized agents for requirements, data, solution, and code work together to turn a maker's intent into a functional, enterprise business app. The window was about three months, with Private Preview on the PPCC keynote stage in October and Public Preview at Ignite in November.
My role went beyond owning a slice of UX. I led design, content, and user research strategy for a team of eight designers, and I was accountable for the coherence of the end-to-end experience: aligning designers across parallel workstreams, connecting design direction to Product and Engineering realities, and helping a brand-new team find the rhythm and confidence to ship together.
The core challenge was ambiguity at speed. The organization had shifted to a start-up mentality of shipping fast and learning in production. Every decision sat somewhere between the deadline and the quality bar.
Four things were in motion at once.
Ownership across a matrixed Design, PM, and Engineering organization was still being worked out after the reorg. We'd previously relied on formal review gates. The pressure of the moment called for something far more agile and fluid, and nobody had built that muscle together yet.
We had to decide what the experience was while building it. Information architecture became the crux. We landed on three primary buttons: Plan, Data, and App, and how a maker moved between them shaped everything else. I led the information architecture myself as a super IC. It took many rounds of iteration, backed by rapid user research, and what shipped was welcomed by customers and senior leadership alike.
We were learning in real time what generative models and multi-agent orchestration could reliably do. Design couldn't stop at the interface. We had to understand model behavior closely enough to know which experiences we could promise customers and which needed to wait.
AI changed what quality meant. A beautifully designed interface could still produce an inconsistent generated app, so "good" had to cover far more than the screens we drew.
A beautiful interface covered two of them. The other five lived in what the agents generated and how they behaved.
Early evaluations were candid about the gap between where generated apps were and where they needed to be.
So Design leaned in across the end-to-end model evaluation process, taking a holistic, multi-pronged approach to reach the highest model quality we could. Before output, design taste briefs shaped what the models generated. After output, designers ran human reviews of generated apps and established research-backed UX evaluation criteria to improve model output. We partnered closely with Engineering to build internal grading tools too, including side-by-side comparisons of generated apps. Together, this moved the conversation about model quality from opinion to evidence.
We weren't just designing screens. We were helping define what good meant for an AI system whose orchestration, outputs, reliability, and UX all had to work as one.
Rather than trying to solve every problem myself, I focused on creating clarity at every level. Four moves became the operating system behind the product.
I used design storytelling and tangible prototypes to align cross-disciplinary partners on one end-to-end vision, which gave each workstream a larger narrative to build toward. Presenting it to senior leadership early secured buy-in that made collaboration and execution smoother across Engineering and Product. Inside the team, I led a values-based brainstorm and co-created a team mission statement, so we had shared principles to hold onto when the pace got hard.
I split the team into squads, each owning an area of the end-to-end, and set up triad structures that mapped responsibilities across our Design, PM, and Engineering partners. I traded heavyweight gates for an agile rhythm of weekly cadences, including big-picture syncs, wall reviews, quality reviews, and pair-coding sessions. People could move quickly without stepping on each other.
We moved away from sequential handoffs. Designers paired in real time with the engineer coding their design, getting pixel-level details right in production code and removing the need for most formal design reviews. Using GitHub Copilot, engineers generated production code directly from design mockups, shortening the dev cycle.
I also mentored the team into designing with generative AI. Code-based prototypes showed animation and interaction far better than static frames ever could, and they compressed the path from design to production. The bigger shift was cultural: the two disciplines started solving problems together.
Every week, I ran a quality review: a cross-discipline bug bash with most of the design team alongside partners from Product and Engineering, focused on overall product quality. Together we went through the product end to end and examined the UX, UI, content, and interactions, so that whatever we shipped to production met a high quality bar. It made quality a shared habit across disciplines rather than a final gate owned by one reviewer.
The whole product, end to end, through four lenses, so whatever reached production met the bar.
About two months after the team formed, the new experience debuted on the PPCC 2025 keynote stage and opened to customers through an Early Access Program. Demand was immediate, and customers described the experience as polished from the first session.
Private Preview wasn't the finish line. We treated it as a learning loop, with a dedicated partner cohort and structured follow-up sessions feeding each round of design and model work.
Moving toward Public Preview meant facing hard reliability realities head-on and fixing what mattered most before Ignite. In parallel, we stood up vibe.powerapps.com as a standalone surface, from identity and deployment to compliance.
“Power Apps doesn't just know what you want to build, it understands why.”
Ryan Cunningham, Microsoft Power Platform Blog, Nov 18, 2025
In about three months, a newly formed team went from a blank canvas to a product in thousands of customers' hands. The practices outlasted the launch, too: AI-first design, model evaluation, and design-engineering pairing became how the team kept working.
Starting from nothing taught me that on AI-native products, much of the leadership happens between the workstreams. It means turning ambiguity into ownership, linking vision to detailed execution, making sure separate features add up to one experience, protecting craft while keeping pace, and building ways to measure model quality instead of debating it.
It also changed how I see Design's role. Building AI experiences pulls designers upstream into product framing and downstream into model behavior, evaluation, and implementation. Design can contribute more than artifacts. It can supply the connective system that helps a multidisciplinary organization move as one.
Most of all, it confirmed the kind of leader I want to keep becoming: one who creates clarity without shutting down exploration, drives urgency without losing humanity, and holds a high quality bar while giving others room to lead.
My biggest contribution wasn't a screen. It was the system around the work,the connective tissue that let a new team ship as one.
Full case study · Password protectedRead the complete story, with the metrics and prototypes ↗