When Making Gets Cheap, Judgment Becomes the Craft
Product
Copilot Managed Runtime
Role
Super IC · Vision, 0-1
Shipped
Sep 2026Public Preview
Context
A new front door, and a new kind of design process
The Managed Apps homepage redesign began as a focused effort to establish a new product identity for managedapps.cloud.microsoft.com, but it evolved into an exploration of how AI is reshaping the role of the designer.
Throughout the project, AI accelerated exploration and expanded the solution space. What emerged was a deeper examination of how taste, judgment, and technology intersect throughout the design process. Designing in code helped transform abstract possibilities into tangible experiences that could be quickly evaluated and refined, while strategic thinking, design judgment, and cross-functional alignment remained essential to driving quality outcomes.
Despite its relatively contained scope, the work evolved through multiple strategic pivots and close collaboration across teams. In doing so, it revealed a meaningful shift in both process and mindset, demonstrating an emerging design dynamic: one where the designer's role increasingly moves from creating every artifact by hand to curating, directing, and refining possibilities with clarity, intent, and taste.
Tension
Speed moved the bottleneck
The stakes were strategic before they were visual. This homepage had to become the front door for Microsoft Copilot Managed Runtime, and it had to earn its place as a distinct experience inside a product family that was becoming more unified.
Many people had a stake in it, and each had a different picture of "great." Product partners wanted to ship quickly and stay aligned with established patterns. Design partners wanted something future-forward enough to earn the exception. Every pivot in the project was the direction swinging between those two poles.
Pole AProductShip fast. Align with established patterns.
Pole BDesignPush the boundaries. Be future-forward enough to earn the exception.
Every pivot swung between these poles
AI-assisted prototyping let me build a working version of each direction in hours instead of days. That should have made agreement easier. Instead, it surfaced five frictions none of my previous projects had prepared me for.
In Figma, exploration has a visible shape: frames sit side by side and directions can be compared at a glance. Code has no canvas. Each variant was a complete working experience, but it lived in its own branch or folder, so comparing directions meant relying on memory. That quietly favored whichever version I'd seen most recently. My fix was to give every variant its own address inside a single build: adding a hash shortcut to the end of the URL (#home-v1, #home-v2, and so on) jumped straight to that direction. Anyone could jump between directions, and stakeholders could compare options in adjacent tabs on their own, without relying on my descriptions or waiting for me to share my screen.
Figma is a room people walk into. Code is a desk one person sits at. Partners who would normally collaborate on a frame or work through design problems in real time were now reviewing finished-looking builds they couldn't touch. Feedback arrived later, and it came as reactions rather than contributions. The work got faster and lonelier at the same time. I deliberately rebuilt the shared moments the old tools provided for free, and it still wasn't enough. The tools were changing faster than our team, and the organization around it, could adapt.
03
Two altitudes, one brain
Design asks what should this become? Code asks why is this button two pixels off? Switching between them wasn't just a context switch; it was a change in altitude. An afternoon spent fixing a responsive layout could quietly erase days of thinking about the product's identity. Implementation problems are concrete and solvable, so they kept crowding out the harder, vaguer questions: where the experience should go, which edge cases actually mattered, and how one landing page could serve a user journey spanning the broader ecosystem. The button always got fixed. The bigger questions just kept waiting.
04
Tangibility changed minds
A static mock could earn a nod in review, but the coded version of the same design sometimes reversed that decision because it felt different once people used it.
With so many competing definitions of product excellence, the mock had been doing something subtle: it let each stakeholder see the version they had imagined. The prototype showed everyone the same thing.
That made the conversations more honest, and it had a real cost. Engineers were building the production experience while I was still designing and coding it, so every pivot a prototype triggered sent them back to the start. The cheaper it became for me to change direction, the more expensive each change became for the people building it.
A working prototype only saves rework if the people building from it treat it as part of their workflow, and that adoption has to be designed just as deliberately as the page.
05
Speed outran judgment
These frictions added up to the real problem. When a new direction costs almost nothing to generate, the list of things to evaluate grows faster than the time to evaluate them. Synthesis, alignment, and long-range thinking don't produce an artifact, so they were the first things squeezed out.
The challenge was never generating ideas. It was creating enough space to evaluate them deeply, align around a coherent direction, and think across the end-to-end journey, not just the screen in front of me.
Turning Point
Choosing the medium by the question
The question was never Figma versus code. The frictions eased once I chose the medium based on the question I needed to answer.
FigmaCodeNeither
01What are all the directions this could go?Figma
02Is the team aligned on a direction, with buy-in to ship?Figma
03Does it feel right once people use it?Code
04How should it move and respond?Code
05Does it hold up with real content and edge cases?Code
06What does the end-to-end look like, and what did we agree to ship?Figma
07Is this the right direction for the product?Both inform it; neither decides it. That takes user research, customer feedback, and conversation.
The pattern tells its own story: Figma brackets the process at the start and end, and code tests the middle. Motion has its own row because it's where static mocks fall short most visibly, and it's what most often changed stakeholders' minds.
Resolution
Shipped, and a workflow that outlasted it
The redesigned homepage launched with Microsoft Copilot Managed Runtime in Public Preview in September 2026, when Microsoft announced the runtime alongside the new Copilot.
“Microsoft operates the platform; your organization sets the boundaries; you remain in control.”
Microsoft Copilot Blog, Sep 25, 2026
The workflow outlasted the project. Teammates adopted the same hash-based versioning in their own coded prototypes, and prototype versions moved into shared team infrastructure instead of living in one designer's personal setup.
Reflection
Where the craft went
With this many definitions of great in the room, my job wasn't to win the argument. It was to make each definition tangible enough to be judged on equal terms. AI made that cheap. What it couldn't do was decide which versions deserved to become real, sense that a direction was right before anyone could say why, or know when to stop exploring and commit. That's judgment, taste, and discernment, and it stayed squarely with me.
The rework engineering absorbed on this project became my next problem to solve. I've since partnered with our team's Design Engineer on a production-ready design engineering pipeline, so what designers build in code can ship without engineering rewriting it. We're also building production-grade templates that keep AI-built prototypes on the rails from the first prompt. The goal isn't to turn designers into engineers. It's to make what designers build worth keeping. When design's work ships instead of being handed off and rebuilt, the discipline stops describing the product and starts shaping it directly, and its influence grows with that.
The lesson wasn't how to design with AI. Anyone can build the screen now.The job is knowing which one deserves to ship.
