{{ activeCase.tag }}
{{ activeCase.title }}
{{ activeCase.summary }}
{{ t.cd_l_tldr }}
- {{ td.l }}
- {{ td.d }}
{{ activeCase.overview }}
{{ activeCase.challenge }}
{{ ap.t }}
{{ ap.d }}
Redesigned end to end, with the in-house studio
I worked with SysAid's in-house studio, led by Tomer Gilat, alongside the product manager, engineering and our design partners, who joined interviews, live sessions and prototype tests. Together we ran competitor and user research with qualitative and quantitative interviews, then took the queue and ticket page from wireframes to tested prototypes. What came out was a design system and an entirely new end-user UI.
Read as text
- Context: Classic, the legacy platform in use since 2002 → Spaces.
- Who: me, Senior UI/UX Designer · SysAid's in-house studio, led by Tomer Gilat · product manager · design partners · engineering and senior stakeholders.
- 1 Research: competitor and user research, design partners, qualitative and quantitative interviews. 18 interviews · 240+ surveys · 6 competitors.
- 2 Synthesis: 3 recurring themes · 4 segments · 4 principles.
- 3 Wireframes: the queue as a clear table with fast filtering · the ticket page as a labeled, sectioned layout.
- 4 Prototypes and testing: 3 usability rounds on service-record resolution, related items and action-item workflows.
- 5 Delivered: a design system · an entirely new end-user UI · Template Designer.
Sources: the case text and Daniel's notes. No invented data.
{{ activeCase.researchNote }}
Read as text
- Methods: 18 user interviews (admins and agents) · 240+ survey responses · 6 competitors benchmarked · 3 usability testing rounds.
- Who: SysAid's in-house studio (led by Tomer Gilat), the PM and design partners.
- Design partners (interviews, live sessions, prototype tests). Valued: the reliability of Classic. Pain: complex UI · limited customization · clunky asset management.
- Customer surveys. Valued: customization · API integration · ticket tracking. Pain: customization complexity · unresolved tickets.
- Competitors (ServiceNow, Jira and 4 more). Valued: strong customization and dashboards. Pain: overwhelming UI · steep learning curves, industry-wide.
- Segments: small existing, need simple workflows, struggle with the outdated UI and migration · large existing, need advanced customization, face workflow strain · small new, want intuitive design and quick onboarding · large new, demand scalability, wary of steep learning curves.
- Recurring themes: the queue was overwhelming · bulk work was slow · non-technical teams could not shape their own templates.
Sources: the case text and Daniel's notes. No invented data.
{{ rp.t }}
{{ rp.d }}
{{ pe.d }}
{{ pr2.d }}
Every fix traced to a problem
The research pointed at four clear problems with the legacy queue. I designed a deliberate answer to each, and each answer shipped as a feature.
Read as text
- Row 1. Theme: the queue was overwhelming. Problem: an outdated interface, dense tables and text that made it hard to spot what was urgent. Move: a modern queue with chips, avatars and dynamic cues. Shipped: graphic chips. Shift: not stated.
- Row 2. Theme: bulk work was slow. Problem: losing to competitors with a dated look. Move: time-saving power, with inline editing and bulk actions. Shipped: inline edit, every cell edits in a single click. Shift: editing a ticket field, 6 clicks → 1 click.
- Row 3. Problem: a steep learning curve. Move: personal queue views to add, reorder and save columns. Shipped: custom columns and queue views. Shift: queue columns, fixed → configurable.
- Row 4. Theme: non-technical teams could not shape their own templates. Problem: dependence on support, where small changes needed pro services. Move: self-sufficiency. Shipped: Template Designer, no-code. Shift: building a ticket template, pro services → self-serve.
- Foundation, across all rows: one governed design system. Shift: UI components, ad-hoc → 50+ governed.
Sources: the case text and Daniel's notes. No invented data.
The main ticket table
The heart of SysAid. I rebuilt it around scannability and speed - clearer hierarchy, status and priority cues, bulk actions and multi-select - so agents move through queues with far less friction.
{{ activeCase.wireNote }}
- Priority: plain text became colored chips that read at a glance.
- Due date: time left shows as a bar, so urgent tickets stand out.
- Status: labeled chips replace plain text, so a ticket's state is quick to scan.
{{ qf.d }}
Ticket page & insight panel
I rebuilt the single ticket into a focused two-column view with a contextual insight panel (open SRs, relevant KB, time-to-repair) and a clear activity journey - plus complex reusable components like the multi-select assignee.
{{ tt.d }}
The design system
I created and maintained the design system from scratch - color, typography, an icon set and complex components with every state documented - so the whole platform stays consistent and ships faster.
What changed for admins
{{ activeCase.metricsNote }}
{{ mt.d }}
{{ activeCase.challengeTitle }}
{{ activeCase.challenge }}
The system they had
This is the interface operators actually worked in before we arrived - a dense grid of near-identical panels, color-coded state, and repeated controls with no hierarchy or search. Everything competed for attention at once. It set the bar the new system had to beat, and it is the reason a flexible, no-code, search-first console mattered so much.
Research without access to users
Classified meant no interviews, no analytics, no recordings and no way to benchmark competitors. So I worked through the people closest to the system. The Product Owner was effectively its super-user, who built and operated the screens and brought insight from the field. With the Product Owner and the development team I ran recurring on-site sessions at the client's offices, gathered as much internal client data as I could and mapped the commands operators fire most, the failures they chase daily and how critical uptime is to the whole operation. Then I iterated, by trial and error, until one design served every user level. Every decision traced back to that picture.
Read as text
- The frame, a classified program: no access to end users · no analytics · no remote sessions · competitor research blocked, because systems like this are classified.
- Outside the frame: operators, reached through the Product Owner.
- People inside: the Product Owner, the system's super-user, who built and operated the screens and brought insight from the field · the development team.
- Methods inside: on-site working sessions at the client's offices · internal client data, as much as could be gathered · a legacy system audit · a frequent-actions inventory of commands, checks and calibrations, ranked by frequency and criticality.
- The loop: iterative trial and error across different user levels · a Concept Design Review with the client and development · every iteration reviewed against real usage.
- Note: all manual, a pre-AI workflow.
Sources: the case text and Daniel's notes. No invented data.
From daily work to what shipped
With no analytics to lean on, prioritization came straight from the ranked frequent-actions inventory. Commands, checks and calibrations, ranked by frequency and criticality, became requirements, principles and a prioritized plan. Below, each daily need sits next to the answer that shipped and the shift it made.
Read as text
- Ranked by frequency and criticality → requirements, principles and a prioritized plan.
- Reaching a screen. Answer: tabs, where every new tab opens on search plus the common pages. Shift: menu digging → tab + search.
- Firing a frequent command. Answer: quick actions on the common pages. Shift: seconds, not menus.
- Running checks and calibrations. Answer: macros, step by step, with live progress and failure drill-down. Shift: expert-only → guided macros.
- Chasing failures. Answer: one real-time console with a live status panel. Shift: scattered → one console.
- Building a new operator screen. Answer: UI Builder, dragging elements from the ICD, no code. Shift: dev task → drag and drop.
- Across all five, different user levels: operators with 20+ years kept the dense layout of about 60 elements, and hierarchy and search were added instead.
Sources: the case text and Daniel's notes. No invented data.
Two concepts, one validated direction
Before committing to build, I ran a CDR - a Concept Design Review with the client and development - to validate a direction both operationally and technically. I brought two concepts to the table.
Read as text
- Reviewed with the client and development, operationally and technically.
- Concept A, tabbed workspace, selected. Navigation: browser-like tabs, each a full screen of operational elements. Real-time operation: built for it. Speed and feel: the least time and fewest clicks possible, fast and familiar. Outcome: earned technical and practical validation and became the operator console.
- Concept B, aircraft hologram, explored. Navigation: a 3D model of the aircraft as the navigation surface. Real-time operation: never gave real comfort for it. Speed and feel: looked striking, but more gimmick than tool. Outcome: set aside.
Sources: the case text and Daniel's notes. No invented data.
The client and development loved the tabbed workspace, and it earned the technical and practical validation to move forward - it became the operator console that follows.
Trade-offs I made
Real programs are full of tension between the ideal design, engineering reality and how people actually work. A few calls I owned:
{{ dc.idea }}
{{ dc.tension }}
{{ dc.call }}
The operator console
Live telemetry streams into configurable sections - sensors, meters, cameras, toggles and commands - while a status panel surfaces connection state and activity failures the moment they happen. Operators see, decide and act on a single screen, at speed.
Automated macros for maintenance & calibration
Routine check and calibration flows run as macros - step-by-step sequences with live progress, remaining time and a drill-down into every failure with its recommended action. Work that used to demand deep technical knowledge became a guided flow operators move through quickly.
Browser-fast navigation
Screens live in tabs, and every new tab opens on search plus the most common pages with their quick actions attached - so reaching any screen or firing a frequent command takes seconds, not menus.
The system behind the system
{{ activeCase.dsIntro }}
{{ df.d }}
Real frames from the IAI design-system and Element Index files in Figma - the reference developers implemented against.
What changed in daily work
{{ activeCase.metricsNote }}
{{ mt.d }}
{{ activeCase.reflection }}
Requirements-led, to a strict brand book
This was a short project, led by requirements rather than a research phase. I worked closely with the PM, the development team and the Clalit studio, to a brand book with specific do's and don'ts. Each requirement below has the design response it got.
Read as text
- At the center: a workshop portal for members of every health plan.
- Audience: every age, every level of digital comfort, diverse abilities. Response: a prominent search field and layered filters up front.
- Catalog: a wide range, held in person or online, from smoking cessation, alcohol withdrawal and weight management to many more. Response: filters for topic, audience, region, language and format, and the essentials on every card.
- Accessibility: strict adherence to accessibility principles. Response: high-contrast modes and accessibility testing throughout.
- Responsive: one experience from mobile to desktop. Response: catalog, filters and registration rebuilt for touch.
- Brand book: a clear brand book with specific do's and don'ts. Response: a portal style guide aligned to it.
- Registration: guide people through registration without friction. Response: one clear path to register.
- Around it: a short project, 3 months from brief to handoff · close collaboration with the PM, the development team and the Clalit studio.
Sources: the case text and Daniel's notes. No invented data.
From discovery to done
{{ activeCase.flowIntro }}
Read as text
- 1 Discovery and catalog: search, filters (topic, audience, region, language, format) and cards with date, time, seats and cost.
- If no online sessions are available: an empty state keeps the member oriented and offers a way back to step 1.
- 2 Workshop page: session details, the session list, a booking summary and a single path to register.
- 3 Sign-in and registration.
- 4 Summary and confirmation: personal details, consent and one clear action.
- 5 Success: an on-brand thank-you with Clalit's mascot.
- The same flow on desktop and mobile.
Sources: the case text and Daniel's notes. No invented data.
A style guide to keep it whole
{{ activeCase.styleNote }}
Built for every screen
{{ activeCase.mobileIntro }}
{{ activeCase.outcomesNote }}
Product to brand, with a studio I built
Scully lets large companies run AI agents inside their ERP and CRM, under their own rules. I designed every surface people see: the back office operators work in, the brand, the website buyers land on and the films that explain it. This case follows the work mostly in order.
It began as a name and a simple S
In Mar 2026, Scully was a name and a simple S. The idea was clear: AI agents doing real work inside the ERP and CRM a company already runs. Nothing a user, a buyer or a developer could look at existed yet. That was the job.
I built a design studio of AI agents, and I make every call
These are my tools, not Scully's product. Scully's AI agents, like Clara, show up later, in the back office.
Being the only designer on a product this wide, I couldn't do every job by hand. So I built my own design studio: specialist AI design agents, each with one role, from UX and UI to motion, sound, copy, QA and development. I direct it. Every task starts as a brief and a plan, and nothing moves until I approve the plan. Studio agents do the work, the QA agent checks it and I sign off the result or send it back. That's how one designer ran research, product, brand and film side by side.
Director
Daniel
Briefs, approves, signs off
Advisor
Creative director
Product and web
- UX
- UI
- Design system
- Web
Brand and visual
- Graphic
- Illustration
- Icons
- Marketing design
Motion and sound
- Motion
- Sound
Copy and strategy
- Copy
- Strategy
Build and QA
- QA
- Development
The loop
- Brief
- Plan
- I approve
- Studio agents execute
- QA
- I sign off
The product pipeline
- Prototype
- MCP into Figma
- Two linked Figma files
- Developers
Map the market first, then pick the gap
abra wasn't entering an empty market, so I mapped it before drawing a screen. My studio made the competitor research and the prototype's user testing fast; the calls stayed mine. Wonderful sells outcomes and hides the machinery. Jeen.ai governs AI agents for IT. Blocks lets anyone build one in plain language. None gives one operator outcomes, spend and retrieval in one place. That gap became the brief. With no live users yet, I based the product's people on its real roles. Later, the website got its own competitor scan before any new page.
Read as text
- Wonderful: sells outcomes and hides the machinery.
- Jeen.ai: an AI operating layer sold to IT.
- Blocks: lets anyone build an agent in plain language.
- The gap: one hides the machine, one governs it for IT, one opens building to everyone. None of them gives one operator outcomes, spend and retrieval in one place.
- Scully's brief: outcomes, spend and retrieval in one place, for one operator. That gap became the brief.
- Method: competitor research, run fast with my design studio.
Sources: the case text and Daniel's notes. No invented data.
A complex system that stays calm to operate
The back office came first, and it's the most complex thing in this case. Operators run a fleet of Scully's AI agents from it: connect them to company systems, build workflows, watch every action and cap every cost. The fleet view opens with outcomes, then the one alert worth opening. Cost got its own page, because an over-budget agent is an incident, not a setting. It took a slot in a short navigation, and it's the call I'd defend hardest. Hebrew is built in, with full RTL. I prototyped it live in Claude Design, every state clickable, and developers get each feature as a numbered script with empty, loading and error states settled.
Under it: Scully-DS
Scully-DS holds 478 variables and 5 color modes, three dark and two light. Every value is a variable binding, so a new mode is a token change, not a repaint. Not everything became a component: drawer and modal bodies differ per feature.
From a simple S to a brand with its own language
With the product taking shape, Scully needed to look and sound like one company. I explored more than 20 logo directions, and management kept the S. So I rebuilt it: redrawn on a 4 px grid, 77 by 72 modules, so every edge lands on the grid. Around it, a cream ground, navy ink and amber as the only accent. Switzer for headlines, Inter for reading. Then Fluid took over the imagery: quiet grounds, each carrying one real piece of the product UI.
One system every surface draws from
The brand needed the same discipline as the product, so it got its own design system, separate from Scully-DS in the back office. Brand DS v1.8.0 holds every value and file: color, type, layout per surface, imagery, motion, print and accessibility, with color pairs rated to WCAG 2.2 AA. I wrote it so a person or an AI tool can use it, and the team builds decks, flyers, email, social and video from it. When the imagery moved to Fluid, every piece followed.
- Components
- 36
- Themes
- Cream and Navy
- Accessibility
- WCAG 2.2 AA
- Version
- Brand DS v1.8.0
| Pair | Ratio | Level |
|---|---|---|
| Ink on creambrand-navy on brand-cream | 16.6:1 | AAA |
| Text body on creamtext-body on brand-cream | 7.1:1 | AAA |
| Accent text on creamaccent-text on brand-cream | 5.3:1 | AA |
| Amber on navybrand-amber on brand-navy | 9.1:1 | AAA |
| White on navybrand-white on brand-navy | 18.7:1 | AAA |
| Text muted on creamtext-muted on brand-cream | 3.3:1 | Large only |
| Amber on creambrand-amber on brand-cream | 1.8:1 | Never text |
A book for the rules, a portal for the team
The rules and wording live in brand book v2.4, cut per reader: an Extended edition of 136 boards, a Slim edition of 31 and a Management version for decisions. Extended and Slim ship as accessible PDF/UA-1, tagged so a screen reader can follow them.
Extended
136 boards · rules and wording
Slim
31 boards · for teams and partners
Management
for decisions
Accessible
PDF/UA-1 · Extended and Slim
The portal
The portal gives the Scully team the slim edition and the files in one place. Book, design system and portal each have one job, and all three carry one version line.
Brand book v2.4 · DS v1.8.0 · Nov 2026
A site an executive gets in 30 seconds
The back office is dense, made for someone who lives in it. scully-ai.com does the opposite job: an executive who will never log in should get it in 30 seconds. So the hero shows operations, not an illustration. Each sector gets its own terms, Scully's AI agents have faces and jobs and Clara, the voice agent, has its own Voice OS page, now live. Every path leads to one action: Book a demo. Underneath: SEO metadata, performance, reduced motion and accessible forms.
See it live: scully-ai.com (opens in a new tab)
Three films, from product tour to booth
Three films, three jobs: the product tour walks one process end to end, the Clara film makes the case to a buyer and the booth film stops people at an event. All three follow one film direction: a protagonist, the camera as narrator, very little text and the real product on screen, with demo data. Clara, a Scully AI agent, is the same AI-generated character in each. My studio's motion and sound agents built picture and mix; I directed every cut and signed off the final one.
What happens in the film
Monday morning. A company counts its AI agents: dozens, each built somewhere else. The AI bill is over budget again, an agent reached into the CRM without approval and every vendor plays by its own rules. Then Clara introduces Scully and shows the real product: budgets and policies per agent, a managed data layer, connectors that expose only approved tools, roles and permissions, a sales agent built without code and the same agent on a website, WhatsApp or a phone line. She closes: "All your agents, from any vendor. One place, under your control. Including me."
Transcript
Hi, I'm Clara, Scully's AI agent. Someone should manage me. Decide what I touch, cap what I spend and stop me when I go over.
It's Monday morning. You count your agents. It's already dozens. Each one built somewhere else. Nobody sees the whole picture.
First, the money. The AI bill just landed and it's over budget. Again. Which agent spent it? On which model? Nobody can say. And nothing stopped it.
Then, the data. An agent reached into your CRM without anyone saying yes. And your core systems? Every agent leans on them, all at once.
And the rules? Agents from every vendor, each playing by its own rules. No single owner. No shared policies. Honestly? If I were you, I'd worry too.
So, meet Scully. One platform to build, connect and manage every agent.
Budget control first. FinOps shows what every agent costs, by model and by tokens. Here's my row. Yes, I'm on there too. You give each agent a monthly budget and policies. My policy? Go over budget, and I'm blocked. Automatically.
Next, a managed data layer. Open a knowledge base and you'll see its schema. For every field, you decide: searchable, filterable or sortable.
Your core systems plug in through Connectors. Each one shows me only the tools you approved. Nothing more.
And enterprise-grade security. Manage users and roles. For each role, you set what it sees, what it edits and where it has no access. Here's mine. Strict, but fair.
Now let's build one together. A new sales agent. Name it, pick a model and a speech mode. Write its instructions in plain language, the way you'd brief a new hire. Connect its tools. Click Deploy, and it's live. No code.
Want it on your website? Picture it with Scully App. Pick the agent, style the chat, paste one line of code. Just the one. And it's already answering your customers.
Or put it on WhatsApp. Or a phone line. Same agent, every channel.
All your agents, from any vendor. One place, under your control. Including me. Book a demo at scully-ai.com
What you see
Monday morning in a contact center: every line ringing, customers on hold. Clara, Scully's voice agent, answers right away, runs the process step by step in the ERP and logs a replacement order. One caller is over the limit, so Clara leaves that decision to a person. She introduces her Scully colleagues, Amit, Noya and Sophie, and ends at scully-ai.com.
Transcript
Monday morning. You know this one. Every line ringing. Every desk busy. Your people dig, screen after screen. The customer holds. And holds. They ring again. From the top. And it's not even ten.
That's where I come in. I'm Clara, Scully's voice agent. I answer right away. Nobody holds. I run your process, step by step, in your ERP. Order replaced. Logged. Done. Every caller in your queue, answered.
Except one. Over the limit? That's your call. I've done the legwork. My Scully colleagues: Amit, Noya and Sophie. Want to hear me work? I'm at scully-ai.com.
How it came together, month by month
The product came first. The brand, the book, the portal and all three films followed in Oct and Nov 2026, on top of the product work.
I start on Scully: a name and a simple S.
Back office: RAG Config Builder, full theming and the product design system.
Back office: a light Snow theme, a keyboard accessibility guard and new error screens.
Logo exploration ends with the S kept. Website competitor scan. Brand design system v1. Clara film begins.
The S redrawn on a 4 px grid, Switzer, brand book v2, the portal and Fluid. Back office workflows rebuilt. Contact-center film, product tour and Clara film v6. Brand book v2.4 as accessible PDF/UA-1, Brand DS v1.8.0.
I estimate I now work about 5x faster than I did before AI, and without it the films would have been out of reach.
Shipped, in use and honest about what isn't measured
The prototype is what abra shows prospects, and development builds from the design system and the feature scripts, not from screenshots. On the brand side, the book ships in three editions, the portal is up for the team and scully-ai.com is live with its Voice OS page. The product isn't in production yet, so there are no adoption or cost figures to report, and I won't invent any.
Mobbin