Dani Weidner Open to Staff roles

Platform / Design systems and ops

Setting the conventions a whole product runs on

Every creation flow had drifted into its own layout, teams argued the same navigation question on every project, and handoff kept failing the same ways. I wrote the standards the product now builds to, built lifecycle gates from an audit of real projects, and gave the roadmap one demand picture to prioritize from.

In one minute

SituationNew surfaces had nothing to build against, so the product felt like several products, and decisions lived in chat threads instead of anywhere engineering could find them.
What I didSet the layout, navigation, and handoff standards, audited handoff failures into three lifecycle gates, and cross-read 340 logged requests into one demand picture.
OutcomeAll three standards are used product-wide, and the gates are newly adopted.

Demo. The workbench convention in use: looping, no audio. Open full screen ↗

ResultProduct-wideLayout, navigation, and handoff standards adopted
Decision3 to 1Divergent layout families consolidated into one workbench
Decision3Lifecycle gates, built from an audit of 5 handoff failure patterns

Decisions

Decision 01 / 05

One layout: the split-screen workbench

Creation flows had drifted into split-screen workbenches, single-column cards, and card grids. I set the split-screen workbench as the standard: a scrollable form on one side and a sticky live preview on the other, with a variant switcher, a summary tab in place of the old review page, and inline errors on any tab with issues.

A shared layout lets a new surface feel familiar on day one.Design principle, layout standardization
BEFORE THE STANDARD INLINE ERROR BEFORE: EVERY FLOW ITS OWN NEW SURFACES HAD NOTHING TO BUILD AGAINST SPLIT-SCREEN CENTERED CARD CARD GRID ...AND MORE ONE STANDARD THE SPLIT-SCREEN WORKBENCH SCROLLABLE FORM ONE SIDE, STICKY LIVE PREVIEW THE OTHER FORM ERRORS INLINE, NOT ON REVIEW VAR A VAR B VAR C LIVE PREVIEW VARIANT SWITCHER PINNED, SUMMARY TAB REPLACES THE OLD REVIEW PAGE ADOPTED PRODUCT-WIDE
Fig. Divergent layouts converged into one split-screen workbench
Decision 02 / 05

A rubric for steppers versus tabs

Steppers for strictly linear flows, tabs for save-as-draft flows where users move between sections freely. Usage data grounded it: one flow forced every user through a multi-step stepper even though 60 percent created a single item.

Teams carry it forward as their default reasoning, so the debate doesn’t restart on every flow.

STEPPER TABS MISFIT FIXED THE RUBRIC ONE QUESTION, ANSWERED PER FLOW A NEW FLOW MUST STEPS HAPPEN IN ORDER? YES NO STEPPER STRICTLY LINEAR, IN ORDER TABS SAVE AS DRAFT, MOVE FREELY THE TEAM BUILDS TO THIS NOW, NO RELITIGATING GROUNDED IN USAGE DATA PICKED THE RULE, NOT TASTE ONE FLOW FORCED EVERYONE THROUGH A MULTI-STEP STEPPER 60 PERCENT CREATED JUST ONE ITEM REBUILT TO THE RUBRIC THE PATH NOW MATCHES THE FLOW
Fig. Steppers for linear flows, tabs for free movement, grounded in usage
Decision 03 / 05

The design file is the source of truth

Every ticket now carries three things: specific affected paths, explicit expected behavior, and a design reference, with the reasoning behind each decision documented alongside the work.

Why: decisions had lived in scattered chat threads, so the same questions came back at every handoff.

SCATTERED SOURCE OF TRUTH TICKET BEFORE: SCATTERED DECISIONS IN THREADS AND VERBAL AGREEMENTS SAME QUESTIONS AT EVERY HANDOFF AMBIGUITY IN, REWORK OUT ONE FILE, EVERY TICKET ANCHORED SO ENGINEERING CAN BUILD WITHOUT COMING BACK TO ASK THE DESIGN FILE DECISIONS + REASONING, DOCUMENTED TICKET PATHS / BEHAVIOR / DESIGN REF TICKET PATHS / BEHAVIOR / DESIGN REF TICKET PATHS / BEHAVIOR / DESIGN REF TICKET PATHS / BEHAVIOR / DESIGN REF THE BAR FOR A GOOD TICKET: SPECIFIC PATHS, EXPECTED BEHAVIOR, A DESIGN REFERENCE
Fig. The design file as source of truth, every ticket anchored to it
Decision 04 / 05

Gates built from an audit, not a template

Design, research, and content kept entering after scope was set. I audited handoff against real projects, from meeting transcripts to tickets to shipped work, and named five recurring failure patterns.

That audit became three gates (discover, plan, build) with entry criteria and named owners, plus a biweekly design and engineering sync where in-flight work is reviewed while it can still change.

FAILURE PATTERN GATE STANDING SYNC THE AUDIT FIVE FAILURE PATTERNS, TRACED TO SOURCES DESIGN ENGAGED TO EXECUTE, NOT SHAPE SCOPE CUT MID-BUILD, NO WAY UPSTREAM CONTENT AS A LATE-STAGE ADD-ON KICKOFFS WITH OPEN QUESTIONS PROCESS DEFINED, NOT ENFORCED THE CHECKPOINTS ENTRY CRITERIA, NAMED OWNERS DISCOVER EVIDENCE IN HAND PLAN SCOPE SIGNED OFF BUILD READY TO SHIP NO KICKOFF WITHOUT DOCUMENTED DESIGN AND RESEARCH THE STANDING SYNC BIWEEKLY DESIGN × ENG · IN-FLIGHT WORK, STILL SHAPEABLE · DEEP DIVES ON DEMAND ALIGNMENT BETWEEN THE GATES, NOT JUST AT THEM BUILT SO THE SAME FAILURES STOP REPEATING
Fig. Five failure patterns audited into three gates, with a standing sync between them
Decision 05 / 05

Give the roadmap one demand picture

Demand arrived through three channels, each with its own bias: a sales request log, a research repository, and meeting notes. I used Claude to code 340 logged requests from an 18-month window into weighted themes, then cross-read them against the other two channels across 20+ enterprise accounts.

Themes confirmed by all three channels became high-confidence priorities. Roles and permissions is the current example.

ALL THREE TWO CHANNELS ONE CHANNEL THREE CHANNELS EACH WITH ITS OWN BIAS SALES REQUEST LOG 340 REQUESTS, 18 MONTHS RESEARCH REPOSITORY MULTI-YEAR STUDY ARCHIVE MEETING NOTES 20+ ENTERPRISE ACCOUNTS SYNTHESIZED SEPARATELY FIRST, SO NO CHANNEL SPEAKS FOR ANOTHER CROSS-READ ONE DEMAND PICTURE CONFIDENCE SET BY AGREEMENT, NOT VOLUME VALIDATED BY ALL THREE CHANNELS HIGH-CONFIDENCE ROADMAP PRIORITIES TWO-CHANNEL THEMES CARRIED FORWARD WITH THEIR CAVEATS SINGLE-CHANNEL SIGNALS HELD FOR VALIDATION, NOT PROMOTED FEEDS ROADMAP PRIORITY THE STANDING REFERENCE FOR WHAT CUSTOMERS ACTUALLY DEMAND
Fig. Three biased channels cross-read into one picture, confidence set by agreement
Next / Case 05
Designing an AI agent that’s honest about what it doesn’t know