Overview
As a design manager, I treat agentic AI as organizational capability, not a personal shortcut. I define how teams run structured discovery, generate reviewable HTML prototypes, and pass critique gates before work reaches stakeholders or engineering.
The goal is velocity with governance: juniors and seniors explore faster, managers coach quality, and accessibility, brand, and member trust stay human-owned.
Tools in the team's stack: Cursor for agentic build and iteration, Open Design for structured discovery and HTML artifacts, Figma for component specs. Agents accelerate scaffolding; design leadership owns hierarchy, accessibility, and ship decisions.
Discover → prototype → critique → handoff loop
Where I use agentic AI
Agents show up across the design lifecycle, always bounded by explicit inputs and review gates.
- Discovery Short forms lock audience, platform, scale, and brand source before generation starts.
- Direction Brand specs or reference sites are extracted into tokens; inferred palettes stay product-appropriate, not generic “AI beige.”
- Prototyping Responsive HTML with working nav, filters, and drawers. This portfolio is an example of that output.
- Critique Checklist plus multi-axis scoring; fix the weakest dimension before shipping.
- Handoff OKLch tokens, component classes, and labeled interaction blocks for engineering.
Workflow in practice
Each phase has a clear contract so agents do not drift into vague brainstorming.
Brief locked first; visible plan second; pixels third
Discover. Agents ask only what the brief left open: platform, constraints, speaker notes for decks, motion preferences. Skipped questions inherit defaults; nothing is re-asked once answered.
Prototype. I start from seed templates (deck framework, mobile frames, case-study CSS) instead of one-off CSS each time. Screen-file-first: homepage, case studies, and platform variants stay separate files.
Critique. P0 items are non-negotiable: no designer-only controls in product UI, no invented metrics, no horizontal scroll at mobile widths. Sub-3/5 scores on hierarchy or restraint trigger another pass.
Handoff. Artifacts ship as self-contained HTML with documented tokens. Same Fondue discipline I enforce at Centene, applied to agent-generated surfaces.
Principles I keep
- Humans own brand, accessibility, and product ethics. Agents propose; designers dispose.
- Honest placeholders beat fake stats; if a number is not sourced, it does not appear.
- One accent per screen, one decisive flourish. Agent output gets edited for restraint.
- Real interactions over static screenshots: tabs, drawers, and filters should work in review.
- Design systems thinking applies to AI builds: tokens, components, and repeatable layouts.
What this enables for teams
- Faster explorationMultiple directions and case-study pages without each designer rebuilding nav and tokens from scratch.
- Clearer reviewsStakeholders interact with real UI: filters, mobile menus, section nav. That reduces misread static mocks.
- Stronger handoffEngineering receives structured CSS and behavior notes aligned with design-system governance.
- Leadership leverageManagers set standards and review gates; designers spend more cycles on judgment, not scaffolding.
Want to go deeper on a specific client workflow or team rollout? Get in touch. Happy to walk through how this maps to your stack.