This is a story about leading, not just designing
This is really a story about leading a design team through the AI transition — and I used a real, high-friction page to prove it rather than theorize it. I'm leading the redesign of the Quote Summary page, where Chubb agents turn quotes into client proposals, and using it to pilot two bets that will define my team's next few years: a Claude-powered prototype-to-Dev workflow, and a redefinition of what each of my three designers does as AI absorbs the routine work. The redesign is still in validation, but the workflow already let us build two fully interactive, responsive, near-production prototypes in 1.5 weeks — work our engineers say would meaningfully speed up their build.
One page. Two bets.
QS is where agents commit or work around the product. I treated its redesign as the lowest-risk, highest-learning place to pilot the two bets that matter most for my team's future — a Claude prototyping-and-handoff workflow, and new specialized roles — while working inside real constraints like a forced design-system migration and partners who hadn't engaged yet.
The conditions I actually worked inside
A compliance deadline I didn't set, dependencies outside my control, two design systems running at once, downstream partners not yet engaged, a key contributor out during testing, and an AI workflow with no playbook. The honest story is how I kept making decisions inside that — not how I avoided it.
What the data said — and what I didn't take at face value
I started with Fullstory behavioral data, turned it into hypotheses, then tested those against agent interviews — and stayed skeptical when the two disagreed. Five agents surfaced three consistent frustrations: finding and trusting underwriter contacts, the inability to edit core quote details, and not being able to find their proposal documents. One key action (the Quote Documents tab) is used by only 40% of agents, even though accessing proposals is the main reason they're on the page.
From thirteen options to two tested prototypes
From the research, my team used Claude to generate 13 mid-fi directions against requirements I wrote. We whiteboarded all 13, then defined two test versions (A and B) built around the agents' top frustrations. We tested both with six agents using fully interactive, responsive prototypes — and I deliberately ran the test before aligning with the domain lead, then ideated her preferred direction in parallel.
The decisions that won't show up in the screens
The work that won't show up in the final screens: a Cyber cross-sell standoff I resolved with a reframe my domain lead called “the correct answer” — now shipping in two phases, with a tactical version due in a week — a fight over named underwriters I lost but turned into structured evidence-gathering, and a stakeholder's design idea I pivoted from challenging to championing once it earned its place.
Who I had to move, and how
The senior skill nobody documents: getting things done through people who don't report to you.
The Claude-to-Dev workflow — proven, not promised
Using our design system's MCP, we built two complete versions of the page in ~1.5 weeks, fully responsive with all interaction states — closer to production than our usual Figma prototypes, which skip states and interactions we don't strictly need for a test. Because we're generating Angular (our framework), the handoff arrives largely coded. Our dev team said it would significantly increase their productivity and saw no downsides beyond losing the annotation notes Figma gives them.
What design becomes when AI absorbs the routine
I used this pilot to test a new operating model for my team: four roles built around what AI can't do. The Director (me) sets vision and alignment; the Architect masters the AI and technical systems; the Builder turns ideas into well-built, responsive prototypes; the Detective goes deeper than what users say. I designed these by predicting how every cross-functional role will change with AI, then anchoring my team to the work that survives.
Where it stands — no overstating
The redesign isn't shipped — we're mid-validation, with the next iteration starting next week. Here's exactly what's validated, what's still forming, and the proxies that already show the pilot working.
The design leadership that compounds in value
One page became the proving ground for the two shifts that define my team's future — a new AI-powered way of working, and a redefined team — navigated inside real organizational constraints. What I took from it: as AI absorbs routine production, a design team's value moves to judgment, context, human insight, and the organizational work of getting good decisions made.