Design leadership for products that can’t afford to be wrong.

Diligent wanted to go all in on AI. Engineering-led pilots had stalled, and the plan was a Word doc of use cases nobody could picture, estimate, or sequence.

I reframed the brief: users shouldn't have to learn to ask AI better questions; the system should already know the question. When a director resigns, the product should know the seat is open, which bylaws apply, and who's qualified, then surface the next step. I stopped writing decks and built a working prototype myself, then put one designer on each use case and held all six to one system across three surfaces: a command center, agents embedded in existing products, and Teams/Slack.

• The CEO, CTO, and BU leaders committed.
• Product could scope and engineering could estimate.
• It became a two-track roadmap of twenty initiatives.
• We showed it onstage at Diligent's April 2026 customer conference.

On 0-to-1 work, is cross-product consistency a hill worth dying on?

Boards is the system of record for corporate governance, used by roughly 70% of the Fortune 1000. Diligent charged a premium the product didn't look like it earned, and any change risked disrupting power users who open it a few times a year.

I took direct ownership: ran the client interviews, owned the information architecture, and chose the visual direction. I made two contrarian calls. I shipped the visual redesign first and held the IA work ~90 days. And I built for the legacy product clients actually used, not the next-gen one, so the modernized old product leapfrogged the new.

• [DECIDE: which outcomes are cleared for public use. See brief §3 Work 2 and §5.]

A better destination doesn't make change easy. How do you get teams to ship early and imperfect when it's never been the norm?

If AI coding agents are going to build UI, the design system has to be something they can read, not a Figma library people forget to open.

I ran Atlas as a product. We shipped a production MCP server that serves tokens, patterns, and components to engineers and AI agents, with Figma-to-code parity on semantic tokens. I set code-level adoption governance: a drift taxonomy and an exception-by-approval model for custom components. When designers left other teams, I routed backfills to Atlas instead of asking for new headcount.

• 85% of engineers use the Atlas MCP server; daily usage is up 20× since launch.
• 88% of calls are agents pulling live Atlas data.
• 58 production-ready components in design and code; 12 native AI patterns shipped in 2026.
• Light and dark themes adopted across 12 products in four business units.

"AI first" was a goal everyone was evaluated on and nobody had defined. Designers didn't know what it meant or how they'd be judged, while the industry was busy announcing design layoffs.

I defined what AI use looks like at every level, from junior to staff. I wrote a triad playbook with product and engineering that treats AI prototypes as hypotheses, not proof, and makes clear when design enters the work. I built a readiness model: six systems (design system, AI infrastructure, workflow and deployment, triad accountability, track record, customer readiness), each rated 0–5, where the weakest sets the ceiling. And I measured the gap instead of assuming it.

• We replaced opinion with a shared scorecard.
• AI gives design teams 50–70% gains upstream (discovery, ideation, first drafts) but close to none in final design and handoff. Net, end to end, it's about 25–30%.
• Gross isn't net, and the last mile is where the gains stop.

Whose infrastructure, remit, and responsibilities have to change once the whole org works this way?

Building prototypes with AI got fast. Sharing them didn't: screenshots pasted into FigJam, nobody sure which link was current, designers pinging stakeholders by hand.

I designed and built VibeSharing on nights and weekends with Claude Code: one version-aware URL per prototype, feedback pinned to versions, stakeholder alerts on new versions, gated external sharing, and a published MCP server (listed in the official MCP registry) so AI coding agents can deploy and pull design-system templates directly.

• 100+ prototypes deployed since launch across multiple product workstreams.
• It's also how I learned what my teams are actually up against when they build with AI.