My AWS Infrastructure Services design team of 18 was responsible for a portfolio that had grown 3x — from 12 to 34+ services — with no headcount growth. To sustain that pace and quality, AI-assisted workflows were essential. The team was already using AI to write personas, synthesize research, and summarize findings, but there was a gap no one had closed: translating Figma designs into working prototypes using vibe coding.
Only two designers were delivering design this way. Most had heard of vibe coding, and those who tried it ran into our internal tool, Kiro, and got stuck — permissions issues, environment conflicts, setup errors that killed momentum before it started. When they did get it running, the output ignored our design system even when they explicitly told it not to.
The blockers were both technical and psychological. Kiro had a steep setup curve, and team members would get partway through, hit a wall — an unresolved permission, a system conflict, an unclear next step — and give up. Some didn't know where to start. One of my direct reports had struggled with it for weeks and had started to question why she needed it at all, and she wasn't alone. The curiosity was fading into avoidance.
"The team didn't need convincing that vibe coding mattered. They needed one clean win to believe they could do it."
If the team could cross that threshold together, they'd be able to validate ideas, communicate vision, and move work forward dramatically faster.
I organized a multi-day team summit focused on how the team works and grows. To shape the agenda, I seeded a few topics and asked the team to add their own, then had everyone vote on what was most pressing — from mining our data warehouse for customer insights, to navigating tough scope-and-bandwidth conversations with stakeholders, to setting up a personal AI assistant for tracking decisions and career growth. I matched each chosen topic to the person best suited to lead it, and I chose to dedicate one full day entirely to vibe coding with a single goal: every person walks out having built a working prototype.
I led the kickoff, the wrap-up, and a working session on the team's new operating model. Every other session was led by the person on the team closest to the work — by design.
During a hands-on session on mining BI data and analyzing it with AI, we found that getting Kiro set up was a much bigger lift than I'd scoped. If those issues carried into vibe-coding day, the team would hit the same walls that had been stopping them for months — and the entire dedicated day would stall before it started. That made the pivot a key on-the-fly call: I reworked the remaining sessions to fold setup into them, clearing the path so the team could focus entirely on building when the dedicated day arrived.
For the vibe-coding workshop itself, I gave two designers a clear vision and a basic outline, then asked them to drive it. We ran a dry run the week before so they had the confidence to facilitate and so I could catch anything off track before it mattered. On the day, the room became collaborative in a way I hadn't fully anticipated — when someone got stuck, a neighbor who'd just solved the same problem would roll their chair over, and people who had been skeptics became ad-hoc coaches.
The workshop covered everything end-to-end: how to prompt Kiro effectively, how to enforce the Cloudscape design system inside a prototype, how to host it, and how to use Git for anyone who wanted to go further.
By the end of the dedicated day, every one of the 18 designers had built a real, shareable prototype. One designer who had never vibe-coded took what she built directly into a design review the next morning and used it to get stakeholder sign-off on a vision for an internal infrastructure incident resolution tool.
The direct report who had spent weeks questioning whether any of this was worth it became one of the team's most vocal advocates. A few weeks after the summit, she co-authored the multiplayer runbook — a guide for teams of two or more designers working together inside the same vibe-coded prototype. She didn't wait to be asked.
Adoption was fast, and over the following 2.5 months we optimized the capability into a durable practice. We ran office hours where early adopters mentored everyone else. I tasked one designer with working out the Figma-to-Kiro-and-back round trip, and when Figma released the tools that made it possible, we had a workflow ready. Another designer built a commenting system directly on prototypes, then extended it so comments could be turned into trackable tickets. I pushed them to take it further — batching changes together rather than making one at a time — to control token usage and increase the efficiency of each iteration cycle. I also worked with a designer to expand our AI scoping tool, connecting it to the codebase, AWS documentation, and service-specific context so it could scope ambiguous, technical requests accurately.
We went from 2 designers delivering design through prototypes to all 18. Today, several teams work directly from live prototypes with Figma reserved for the broader design overview and a clean round-trip workflow between the two. What started as a summit became a cultural shift — AI moved from a curiosity to a core part of how this team does its best work.
