How I use AI to compress the discovery-to-delivery timeline — without sacrificing the rigor, stakeholder alignment, or craft that enterprise-grade work demands.
After 20+ years building products at Prudential Financial and Northwell Health, the bottleneck was never ideation. It was compression — the time between a clear brief and a deployable, stakeholder-approved, developer-ready design.
AI changed that equation. The strategic thinking, the systems architecture, the stakeholder navigation — that's still human work, and always will be. What AI does is eliminate the dead time: the mockup-then-handoff gap, the research-then-synthesis lag, the iteration-then-review wait.
The result is a workflow where enterprise-grade rigor and startup-speed delivery aren't in tension. You get both. These are three projects that prove it.
Every engagement starts with no prior visual language. I establish the token layer first — color, type, spacing, radius — then build components against it, then build screens against components. Bloomberg DAP produced a 40-component Figma system with token JSON and a React component library. DLC Management produced a cross-format campaign system spanning print, web, and presentation. Crescent Psychotherapy produced a CSS-native system with no Figma file at all. The output format changes. The architecture discipline doesn't.
The DLC "Rent Is Next" campaign is the clearest example: a content audit surfaced the gap between what the client believed and what their materials said. That gap became the campaign brief. From there: whitepaper architecture, data narrative structure, messaging pillars, and a complete visual system deployed across a landing page, executive presentation deck, print-ready whitepaper, and social assets — all from one strategic brief, all anchored in the same design system. AI accelerated the research and synthesis layer; the strategy was entirely human.
Claude Code builds the production experience directly — HTML/CSS/JS, mobile-first, schema markup, Open Graph, accessibility. Antigravity QA's every viewport before any commit lands. The result: production-quality on the first deploy, with no wireframe-then-build phase creating version drift. Crescent Psychotherapy went from brief to live site in weeks. DLC's campaign landing page was built and approved as a hi-fi browser prototype before a single WordPress component was written.
At Prudential Financial and Northwell Health I led teams across design, product, and engineering — building the LIV AI assistant at Prudential (6 CX awards) and enterprise design systems at Northwell still running in production. In the studio model, I play every role simultaneously: strategist, systems designer, visual designer, developer, and stakeholder translator. The AI workflow doesn't replace the cross-functional perspective — it means one person with 20+ years of that perspective can now deliver at the speed of a team.
Three engagements, three scales — a Bloomberg Philanthropies enterprise portal, a full-service campaign for a national real estate operator, and a zero-to-one studio build for a solo practitioner. Same rigor. Different tempo.
Sole UX Lead on a Bloomberg Philanthropies nonprofit accelerator portal. No brief, no brand guide, no design team. Built a complete design system from first principles, ran stakeholder reviews with Bloomberg senior leadership, and delivered a developer-ready Figma package + React component library to a WordPress build team.
The design was presented to Bloomberg Philanthropies senior leadership and approved at the VP level.
Ongoing agency partnership with one of the country's most active open-air retail operators. Built a design system where none existed, then used it to produce a full thought leadership campaign — landing page, executive whitepaper, presentation decks, data visualization, and social assets — all from a single visual language.
The distinctive loop: pages built in Claude Code as hi-fi wireframes, ported to Figma for client review and edits, then rebuilt in Claude Code as production-ready components for WordPress integration.
Zero to live site for a solo LCSW private practice in New York. No Figma. No handoff. No discovery deck. The brief went straight into Claude Code — logo, design system, copy, SEO schema markup, and a production-deployed site built entirely through a conversational AI workflow.
Two AI systems working in parallel: Claude for build, Gemini for research and schema validation. Weeks, not months.
Every engagement starts with a structured intake — not a questionnaire, a conversation. What does the client actually need? Who are they talking to? What does success feel like to them? What is the one thing this can never be?
That conversation gets structured into a working brief: audience emotional state, hard constraints, personality parameters, and a core differentiator. The brief is the anchor. Everything downstream — color, type, layout, copy — is measured against it.
With AI in the loop, the brief also becomes the first prompt. It's fed directly into Claude, establishing the decision framework that governs the entire project. No mood board theater. No discovery-for-discovery's-sake. Brief to build.
The project began with no formal brief — just a program, a program manager (Dave Ebert, Lapine Group), and a set of working sessions. My job was to shape the UX architecture from conversations alone.
Early discovery sessions defined the program structure, the phase logic, and the full page inventory. That oral information was synthesized into a written IA structure that became the nav system foundation and the persistent phase architecture central to the dashboard.
A written brief: practice philosophy, audience, what the site needed to do, and one hard constraint — nothing could feel clinical. That brief went straight into Claude Code. From that moment, it anchored every decision that followed.
"The brief became the prompt — fed directly to Claude to anchor every decision that followed. No mood board. No discovery deck." — Process documentation, Crescent Psychotherapy
DLC leadership arrived with exactly the right kind of brief: they knew their portfolio outperformed. They had the data. They just needed it to look like they meant it. A content audit and strategy sessions surfaced the gap between what DLC believed and what their materials communicated.
That audit became the campaign brief — defining the "undeniable" standard every deliverable had to hit, and establishing the thought leadership positioning that would run from the whitepaper to the landing page to the exec deck to social.
A brief defines the problem. A plan defines the structure. The gap between the two is where most engagements lose time — weeks spent building discovery decks that no one reads, producing wireframes that get scrapped at the first stakeholder call.
The AI-assisted workflow collapses this gap. Once the brief is established, AI helps structure the information architecture, identify navigation patterns, surface the right questions for stakeholders, and model alternative approaches — quickly enough that the plan arrives in the same conversation window as the brief.
The plan isn't handed to the client — it's built with them, in real time, through a series of targeted working sessions. Decisions are documented, rationale is preserved, and the IA becomes the north star for every subsequent design decision.
After early discovery, a structured IA outline defined the nav hierarchy: Home, Your DAP Program (five phase sub-pages), Program Calendar, Directory, and Supporting Materials. That structure shaped the persistent phase-tab architecture that became the visual centerpiece of the dashboard.
Key strategic call: phase pages were elevated from tabbed navigation to full persistent pages — each with its own URL, content stack, and right-rail sidebar. This was an architectural decision made in the plan phase that fundamentally defined the experience.
The plan phase was compressed into the brief itself. Single-page architecture — sticky nav, hero, services, testimonials, booking, CTA — decided immediately from the brief constraints. No wireframe phase. No separate IA document.
Claude structured the component order based on the audience's decision-making journey: trust signal (hero) → credibility (services) → social proof (testimonials) → action (booking). The plan was the prompt.
The content strategy phase built the full campaign architecture — campaign theme ("The Rent Is Next"), whitepaper outline and chapter structure, key data narratives, and the messaging pillars that would thread through every touchpoint. Landing page, whitepaper, executive deck, and social assets were planned as a cohesive system from day one, not as a pile of one-offs.
The plan phase defined what formats were needed, what each needed to accomplish for its specific audience, and how the design system would bridge them.
Enterprise design starts with systems, not screens. A design without a system is a one-off — impossible to scale, impossible to hand off, impossible to maintain. The system is the deliverable; the screens are evidence that the system works.
The AI workflow changes how systems are built, not whether they are. Tokens — color, type, spacing, radius — are defined first, documented as CSS custom properties or JSON, and applied consistently from component to component. AI accelerates iteration on the token layer without introducing inconsistency downstream.
The output looks different depending on the engagement: a 40-component Figma library with React code for Bloomberg, a CSS token block with a living markdown reference for Crescent. The process is the same. The scale adjusts.
There was no prior design system. Built from scratch in Figma — 40+ components, complete token coverage — with Claude Code generating the React component library and token JSON in parallel. Figma Make accelerated the component scaffolding phase, translating Figma frames into code stubs that Claude Code then refined into production-ready components with full state coverage and token binding.
Three output formats delivered simultaneously: machine-readable token JSON, a written reference document, and the production React library — all synchronized to the same source tokens.
"This is truly a thing of beauty." — Dave Ebert, Principal Consultant · Lapine Group, on first seeing the design system playground
No Figma file was opened. The design system was CSS custom properties from the start — a 20-token color palette, type scale, and component library built directly in code and documented in a living markdown reference. The code was the design system.
The SVG logo mark was hand-crafted in code: three paths, two sparkle forms, one crescent body — fully scalable from 16px favicon to full-bleed hero. No vector application needed.
DLC had no unified visual language for campaign work — no design system, no documented color usage, no type hierarchy, no data visualization standards. All of that was built from the brief up: a primary/secondary color system, typography hierarchy for print and screen, layout grids for whitepaper and deck formats, and a data visualization template library.
Every chart was designed to function as an argument, not decoration. The system was built to produce consistent output across landing page, whitepaper, exec deck, and social assets.
Iteration speed is a product of system fidelity. When the design is token-driven and AI-assisted, revision requests that would take a junior designer a day take hours. The constraint is comprehension, not execution.
The AI-assisted iteration loop: receive feedback → interpret it against the brief constraints → identify what in the system changes → execute the change → validate against the system. That cycle runs in hours, not days. Stakeholders see revisions the same morning they're requested.
This speed creates a different kind of stakeholder relationship — one where feedback cycles feel collaborative rather than transactional. The client stops managing a timeline and starts doing design work alongside you.
Bloomberg-side feedback came through Dave Ebert as a primary relay point. Feedback from one review session was regularly incorporated and turned around same-day or next morning — hero image treatment, working group peer display, program calendar copy, navigation restructure. Multiple structural changes, implemented in hours.
The Figma comment thread became the asynchronous revision log — Dave would tag components directly, I would resolve and confirm. No revision documents. No PDF markups. Living design, living feedback.
Because the build environment was live from the first commit, iteration happened in the browser. Every change to the token system immediately propagated across all components. Color adjustments, type scale tweaks, spacing refinements — all reviewed in context, not in a mockup.
Claude iterated on path geometry and color relationships in VS Code; Antigravity previewed each pass in real time. The feedback loop was minutes, not days.
The landing page was built in Claude Code as a high-fidelity wireframe — fully styled, responsive, and interactive — before any approval conversation happened. Instead of presenting grey-box mockups, the client reviewed a live browser URL that looked and behaved exactly like the final product.
Antigravity handled live preview throughout. By the time the client saw the page, it was already production-quality. The approval was "yes, ship it," not "now go build the real thing."
Testing is not a phase that comes after design. In the AI workflow, testing starts at the first commit. Every component is reviewed in context — in the actual viewport, on the actual device — before it is considered done.
For web builds: Antigravity handles live browser preview and cross-viewport QA. Layout breaks, carousel behavior, mobile nav states, scroll triggers — all caught before any push. For structured data and SEO: Gemini validates schema against current spec. For enterprise portal UX: Figma prototype flows demonstrate interaction logic to stakeholders without code required.
The goal is that by the time a design reaches a developer, every interaction state, every responsive behavior, and every edge case has been observed and resolved. Zero surprises in the build phase.
Figma prototype flows demonstrated the phase-tab navigation, the document card expansion pattern, and the directory cross-link behavior — all reviewed by Dave and the Lapine Group before any WordPress development began. The prototypes were the test environment.
A Loom screen walkthrough was produced specifically to help the Bloomberg team and developer navigate the interaction logic visually — a practical QA artifact that eliminated questions during the build phase.
Antigravity caught every layout break, carousel state, and mobile nav issue before each commit. Schema.org MedicalBusiness markup was validated by Gemini against current specification — services array, credential markup, booking action schema, sameAs cross-links — before the first deploy.
The result: no post-launch schema errors, no mobile regressions, no layout surprises. Production-quality on the first push.
Enterprise design doesn't live or die on craft — it lives or dies on alignment. The ability to translate design decisions into business language, navigate proxy feedback chains, and present work at a VP or C-level without it getting shredded in the room: that's the senior designer skill that AI doesn't replace.
The AI workflow changes the setup conditions for stakeholder review. When the design arrives backed by a documented system, clear design rationale, and a Loom walkthrough of the logic, stakeholders spend the meeting responding to the work instead of orienting to it. That changes the quality of the feedback — and the quality of the alignment.
The workflow also supports asynchronous review: Figma comment threads, annotated screens, and decision logs mean that stakeholders can review at 10pm and have responses waiting when they open their laptop in the morning.
The design went to Bloomberg Philanthropies senior leadership on March 20, 2026. It was presented by Dave Ebert, who had already gone through one full design review round with my annotations and a Loom walkthrough to prepare. Dave's message after the final delivery:
"LOOKS AWESOME!! Thank you! I am presenting these to BP on Friday — wish me luck!!" — Dave Ebert, Principal Consultant · Lapine Group
The designs passed the Bloomberg leadership review without structural changes. That's the outcome of arriving prepared.
Most enterprise engagements involve a proxy stakeholder — someone translating client feedback before it reaches the designer. The workflow accounts for this: design decisions are documented with rationale that helps the proxy stakeholder advocate for the work in rooms where I'm not present.
Bloomberg confirmed the value of document status tracking via a forwarded email thread — a simple validation that proved the right-rail checklist approach before it was ever debated in a design review.
Most tools work in one direction. The DLC engagement required a workflow that moves fluidly between live code and Figma — building in Claude Code, reviewing in Figma, editing with the client, then extracting production-ready components back out for WordPress. This loop is only possible when you own both sides of the process.
The landing page is built as a high-fidelity working prototype in Claude Code — real HTML/CSS, responsive, fully styled against the design system. Antigravity previews every change in the browser. When the build is approval-ready, the design is ported into Figma where the client can review sections, request edits, and annotate directly. Revised components are then rebuilt as production code in Claude Code for handoff to the WordPress development team. The same visual language, the same tokens — two environments, zero drift.
The highest-value moment in any engagement isn't when a stakeholder approves a design — it's when a designer prevents the wrong design from being built. Scope management, simplicity advocacy, and the ability to say "I hear what you're asking for, and here's why a different approach serves your users better" — that's leadership, not just design.
In the AI workflow, the research layer validates these positions. When a stakeholder proposes adding complexity, AI helps rapidly assess the implementation cost, the user impact, and the maintenance burden — and makes the tradeoff conversation concrete, not theoretical.
The discipline is knowing when to push back, when to defer, and when to prototype both options and let the evidence decide. 20+ years at Prudential and Northwell teaches you to read those moments.
LLM search: Dave proposed an AI-indexed search against 400+ files. After the dev lead flagged implementation complexity and I validated the maintenance risk, we scoped it as a future enhancement and proceeded with conventional filter-based search. Right call. Simpler launch, no maintenance debt.
Nav dropdowns: Dave initially wanted dropdown menus. As the full site structure clarified, I held the position that dropdowns were unnecessary. After seeing the final design: "I hate em, so let's lose all of the drop-downs, you were right."
Document tray: Dave proposed a persistent right-column document tray. Recommendation: load curated default state, apply search on demand. Dave deferred to the simpler approach.
The hard constraint — "nothing clinical" — was the most important design decision on the project. It was established in the brief and never revisited. Every visual choice, every line of copy, every layout decision was filtered through it.
Gemini pressure-tested the copy against HIPAA-adjacent considerations and validated that the schema markup didn't inadvertently signal clinical authority in ways that could mislead vulnerable users. Research in service of constraint enforcement.
Design doesn't end at the Figma file. The final 20% of a project — developer integration, QA in the actual build environment, mobile breakpoint resolution — is where craft differentiates from competence. A developer who receives an ambiguous handoff builds an ambiguous product.
The AI workflow ensures that what reaches the developer is complete. Token-driven design systems produce JSON files ready for direct CSS or framework import. Component annotations eliminate interpretation. Loom walkthroughs transmit context that no static document can. The designer stays engaged through the build phase — not to manage, but to ensure fidelity.
For web builds, mobile-first responsive design is the starting point, not a final check. Every layout, every component, every interaction is designed for the smallest viewport first and scaled up — not retrofitted down.
The handoff package to the WordPress build team included: fully annotated Figma file, design token JSON for Token Studio import, and a production-ready React component library. Figma Make was used to generate initial component code stubs directly from the Figma frames; Claude Code then built each component to full production standard with complete state coverage, token binding, and accessibility attributes.
The Figma Make → Claude Code pipeline meant the React library stayed in sync with the Figma source throughout the engagement — no drift between design and code. Go-live April 30, 2026 — on schedule.
There was no handoff. The designer was the developer. Claude Code built the production HTML/CSS/JS directly — sticky nav, animated hero, services grid, testimonial carousel, three-platform booking integration, Calendly CTA. Mobile-responsive from the first commit.
Pushed to Netlify via CNAME on crescentpsychotherapy.com. Full Open Graph, Twitter Cards, canonical, schema.org MedicalBusiness markup. Production-quality on the first deploy.
"Total tools opened: VS Code, a terminal, two browser tabs. Total time from brief to deployed site: weeks, not months." — Process documentation, Crescent Psychotherapy
The same design system produced five distinct deliverable types — landing page HTML/CSS, print-ready whitepaper PDF, executive PPTX deck, animated data visualizations, and social media assets. One visual language. Zero rework between formats.
Complex components built and approved in the code-to-Figma loop were rebuilt in Claude Code as clean, documented components for the WordPress development team. Full campaign suite deployed in time for ICSC Las Vegas 2026. Engagement ongoing.
Three engagements. All delivered on time, at enterprise quality, with zero late-stage surprises. The workflow scales from a solo practice site to a multi-format thought leadership campaign to a Bloomberg Philanthropies program portal — and the output quality doesn't change.
Every engagement moves through the same framework, regardless of scale. The AI tools change the speed of execution. The rigor of each stage doesn't.
Fractional UX leadership. Full-service design work. Either way, you get 20+ years of enterprise-grade thinking — at the speed AI makes possible.