Bridging the Gaps in AI Design

Microsoft's Business Innovation Center (BIC) builds AI capabilities into business applications and Copilot Studio. Its designers were working at a pace their process was never built for: research, validation and user feedback were the first things to go when a deadline moved, and nobody had a shared way to say what “responsible” meant for a feature that was due next sprint.

That gap carried two risks. Features could ship that didn't meet real user needs, and AI behaviour could slip past ethical lines that nobody had time to check. BIC partnered with SCADpro, SCAD's industry collaboration studio, to find a better way, and I was asked to run the project.

Microsoft's goal was ambitious and deliberately open: a responsible, human-centered framework that embeds AI into product development from the start. I ran kickoff and alignment sessions with the BIC stakeholders to find out which of the six deliverables mattered most, and what would prove the work was useful rather than just interesting.

The answer reshaped the project. A new process diagram would be forgotten within a quarter; principles that teams actually used, with evidence that they used them, would not. So the scope became three linked things: principles worth following, a light system that keeps teams honest to them, and a report that proves it to leadership.

I split the research into two tracks run in parallel. Three secondary research pods covered team culture and mindset, current collaboration, AI capability design and principles in practice, each with a lead and a weekly rhythm of findings. Alongside them, a primary research track ran in-depth interviews with the people who live the problem.

We spoke to 34 people across Microsoft and other product teams: designers, PMs, researchers and team leads. Every interview fed one shared affinity board, so the pods were never synthesising in isolation. Three patterns came back again and again.

We wrote eight AI-first design principles, built from the research, Microsoft's Responsible AI guidelines and HCI ethics. Each one is short enough to remember, phrased as something a designer can do, and broken into sub-principles that can be checked against real work.

Principles only change behaviour if they show up during the work. An ideation workshop across all four disciplines produced concepts from synthetic users to research pop-ups, and two were taken to prototype: a physical, rubber-duck style desk companion, and Mictreesoft, a digital tool. We tested them side by side.

Ideation Workshop

A/B Concept Test

With Mictreesoft chosen, I mapped the whole system across a project's life: align on the principles that matter at the start, apply them through light touch-ins in the middle, and reflect on them with evidence at the end. Each phase asks as little of the team as possible.

The Framework

The team brought four disciplines: design technologists, researchers, design strategists and visual designers. I organised them into cross-functional pods with clear owners, ran the project in short agile sprints, and kept a single roadmap that both SCAD and Microsoft could read at a glance.

19 students · 4 disciplines · 1 roadmap

Design StrategistsTurned insights into frameworks and the principle set.

ResearchersRan the interviews, the secondary pods and synthesis.

Visual DesignersMade complex AI concepts clear, from the tree to the report.

Design TechnologistsPrototyped the agents, flows and the working tool.

Rituals kept 21 people moving as one. Daily stand-ups surfaced blockers early, sprint planning and reviews tied each pod's work back to the scope, and regular check-ins with the BIC stakeholders meant there were no surprises at the midpoint or the finish. Retrospectives closed every sprint.

We also agreed how we'd work together. A short team charter set the tone: open minds, actionable feedback focused on the work, accountability for what you own, and respect for each other's time, including no messages after 6 PM. It sounds small, but it kept a large team generous with each other under deadline pressure.

As the team's design technologist, I built Mictreesoft's engine as a crew of three AI agents running on Microsoft's internal models. The first reads a project's context and writes a question of the day, weighted toward the principles the team chose. The second scores how well the team's answers adhere to each principle. The third turns everything into an executive report that ties design behaviour to business outcomes.

The experience is built around a tree. Every project starts as a seed; each honest answer helps it grow, and each principle hangs from its branches with a live adherence score. It turns something abstract into something a team can see, care for and be a little proud of. Try it below.

I presented the final system to Microsoft's BIC stakeholders on May 27, 2025, alongside the team. We showed the principles, a live walkthrough of Mictreesoft on a sample Copilot project, and the executive report it produced.

The Curtain Call

The stakeholders singled out the principles as the novelty they hadn't seen elsewhere, and the executive report as the piece that made ambiguous design work clear, adding proof and accountability where there had only been intent. BIC adopted both the principles and the reporting style. The interaction design behind the tree, a light daily nudge that keeps principles in view while you work, was merged into Copilot and deployed, without the tree.