When I joined amp, its AI Chat was a standalone feature that offered little value beyond basic Q&A and generic workout recommendations.

Despite being positioned as a personal AI coach, it lacked meaningful personalization, access to users' fitness data, and integration with the app's core features.

The experience was also overwhelming. Users were greeted with their entire conversation history rather than clear, actionable starting points.

Before moving into high-fidelity design, I mapped how users would enter the chat and what they could do from there: first-time and returning entry, contextual entry points like logging a meal or starting a mobility assessment, and the full insights and notifications loop.

This map became the shared reference for product and engineering, and the foundation for the final flows.

As the AI Coach evolved into a central product experience, its presence in the app needed to reflect that shift.

I redesigned the chat entry point within the bottom navigation, introducing a visually distinct, almost floating button that separates it from standard navigation items.

This reinforced the AI Coach as a prominent, always-accessible entry point.

Rather than expecting users to know what to ask, I redesigned the chat entry experience around clear, actionable starting points.

Suggested actions such as Log a Meal, See My Insights, Mobility, and Adjust Plan make the coach's capabilities immediately discoverable.

For returning users, a focused previous-conversation preview allows them to resume interactions without displaying the entire chat history.

In the old experience, opening the chat started recording automatically. That makes sense mid-workout, when your hands are busy. But most people opened the chat to get information, and an open microphone was the wrong default.

As a user myself, it felt intrusive. I usually closed the chat right away.

In the new experience, the chat opens on guided starting points, and the input bar works like the AI tools people already know: type, dictate, or switch to voice mode. The user decides, not the app.

The redesigned experience introduces contextual prompts that guide users toward relevant information based on their training activity.

Instead of relying entirely on user-initiated questions, the AI Coach can surface personalized insights and suggest meaningful next steps.

This creates a more useful experience grounded in the user's fitness journey.

Most fitness apps show charts and leave users to work out what they mean. I designed insights to answer three simple questions in plain language, then let users act on the answer right away.

For example, when a user dropped weight mid-set three times in one workout, the coach showed:

• What happened. "You dropped weight mid-set 3 times today."
• What it means. A short, plain explanation of why it matters for their training.
• What you can do. A concrete suggestion, like adjusting the starting weight before each set.

One tap on "Yes, adjust my plan" updates the training plan. No settings to search for and no jumping between screens.

The redesigned AI Coach connects users to meaningful capabilities across the app.

Users can access their training progress, review mobility assessments, log meals, explore personalized insights, and adjust their plans through a single conversational interface.

Mobility results feed straight back into the plan. The training program adapts to the score, and progress is re-checked four weeks later.

The goal was not to replace the rest of the app, but to make its capabilities easier to discover and use.

Once the coach was easy to find and every conversation had a clear starting point, people used it more and went further, without a dedicated launch.

Gloat was built bottom-up. Employees mapped their own path, say from product designer to senior, lead or head of design, and saw which skills they needed to get there. Enterprises used it to retain people and hire from within.

AI changed the question. It's no longer only where one person wants to grow, but how every role in the organization changes when AI takes on part of the work. I joined to lead design for that pivot: a top-down view of every role and every person, built for the people who make workforce decisions.

I worked as a close design partner to our customers, running ongoing interviews with executives, HR leaders and mid-level managers.

The interviews showed how each of them evaluates AI impact, and where trust breaks down. Leaders had invested in AI but couldn't turn it into decisions:

• They couldn't see which tasks and roles AI actually affects.
• They had no reliable way to measure AI value beyond pilots.
• Static job definitions didn't fit the hybrid roles they were already seeing.
• Even with numbers in hand, there was no clear path to an organizational decision.

To test these questions against real work, I flew to Singapore for a three-day, on-site design thinking workshop with Grab.

We started from their actual work processes and mapped how AI reshapes who owns each task. On the whiteboard, the problem split in two. Work redesign: changing the job itself (about 20%) and the process around it (about 80%). And workforce redeployment: reskilling through L&D, redeployment through HR business partners, and org design with workforce planning, finance and HR.

We also mapped who owns the change, from the group level, through an AI center of excellence, down to business partners and every employee.

That insight moved the product from role optimization to process redesign, and from a single dashboard to a connected set of tools. Each tool answers something we heard in the interviews or on the whiteboard.

A few UX principles held it together:

• Overview first, detail on demand. Four headline numbers, then a heatmap by job family, then sorting and drill-down. Leaders get the answer before the analysis.
• Speak the language of the decision. Job families, savings and FTE, the terms leaders already use in planning, rather than model scores.
• Every number explains itself. AI impact always sits next to the tasks, job codes and people behind it, so leaders can trust and defend it.

Job transformation was born in the Grab workshop, as the answer to its hardest question: once you know which roles AI affects most, what do you do with them? Do they get a new role, or take on work from other roles?

It turns the analysis into action, starting from a single role. Leaders pick one of four paths: transition people to another job, create a new job from current tasks and market trends, restructure the tasks and skills of a role, or consolidate several jobs into one. Each result shows which skills carry over and which need to be added, so the plan is concrete from the first screen.

The flow is deliberately short: one search, four clearly named paths, and results that show what carries over before what's missing, so change starts from people's strengths.

Designing for AI at the organizational level is as much about change management as it is about UX.

The project started with a request from the NBA: they needed more unique content, more often.

WSC already produced video at scale, and held a huge, tagged collection of game footage and data. The raw material was never the problem.

The problem was that every piece of content lived in a single format. A video stayed a video, an article stayed an article, and each new format meant starting from scratch.

I ran bi-weekly interviews with NBA and ESPN content teams, so every iteration was tested against how they actually work.

With product and engineering, I mapped the system on the whiteboard first: how an article becomes clips, how clips become a video, and how A2V fits into the wider Studio and its other flow starters.

The flow has just two steps: Source and Preferences. Nothing about video editing is asked up front.

Users write their own story or paste a link to an existing article. A2V fetches the text and drops it straight into the editor.

AI writes the first cut, but the editor decides what matters. Hovering over any part of the text opens a small toolbar to highlight a moment, fix the wording, or remove it, before a single clip is generated.

Shaping the text shapes the video. Editors don't need to learn a new tool to direct the result.

The second step sets how the video looks and feels, split into Video settings, Publish, Graphics and Folders.

• Video type. Play-by-play, image or story, so one article can become different outputs.
• Background music. Options speak the language of the game: chronological, every quarter, every half, every OT.
• Transitions. Start, mid and end transitions, linked with audio by default.

Preview is always one click away, so users can check before they create.

Once the article is converted, the result is a real video, not a locked export. It opens in the same editor WSC users already know, fully editable on the timeline, as if it had always been a video.

The clips list keeps each clip next to the line of the article it illustrates, so editors can scan story and footage side by side, play any moment, and swap or trim what doesn't fit.

Once an article could become a video in a few steps, the same newsroom could publish far more, on every screen.

A2V proved the pattern with its first pair: article to video. The next step was the full one-to-many vision, turning any piece of existing content into the format each channel needs.

And in reverse: take a summary the system already has and turn it quickly into a full article.

When I joined WSC Sports there was no design system. A day in, I told my partner I wasn't going to stay. I stayed three years.

Instead of leaving, I started mapping, on my own, what was actually going on. Each of the five products had grown its own system, never written down and never shared. Every new feature brought its own version of the same components, built again from scratch by a different team.

The mapping made it concrete. The only thing that loosely tied the five products together was color, and even that wasn't the same: the brand color shifted from product to product. In Clipro alone, the flagship product that later became Studio, I found six different buttons and four completely different tooltips.

I mapped every repeated component across all five products, removed the redundancies, and set the shared foundations first: color, typography and the basic elements every screen is built from.

I led the design team on the library while staying hands-on, and led the development team through implementation, so the components in Figma and the components in code stayed the same components. A design system is a living thing: every designer ends up shaping it, so part of the job was setting the rules that keep it coherent.

To lead the rollout with engineering, I built a tracker that mapped every component in every screen: where it lives, which squad owns it, whether it already exists in Storybook, whether it's implemented in the product, and whether it has a Figma design.

It set the order of work and showed each squad exactly what was missing.

The mapping became the brief. The four tooltips became one tooltip family, in dark and light, from a single line to rich content. The six buttons became one button system: primary and secondary, with or without icons, and every state defined.

The same logic carried across popups, dropdowns, fields and filters, so every team built from the same parts.

The most meaningful result went beyond consistency. With a shared language in place, WSC's products came together into one platform, and it changed how the company thought about its product.

Graphics launched in 2017 as a standalone feature for creating branded sports graphics, like player stats, scores and player of the week. In 2019 it became part of WSC Sports' basic plan, so every customer had it.

Having it wasn't the same as using it. The workflow was complex, full adoption took extensive employee training and manual effort, and the product lived in its own corner of the platform, away from where teams actually worked on their content.

I started by mapping the existing flow step by step. Creating one graphic took nine steps across two separate tools. Users could choose only one ratio at a time and had to add parameters by hand, without seeing them update. Clicking Create sent them out to the studio with no way to edit, and changing the ratio meant starting over.

Both personas wanted the same thing in the end: one place to make a quick, on-brand edit without leaving the platform. The current tool failed them in four ways.

If Graphics felt as easy as the tools users were leaving for, and lived inside the product they already used every day, they would have a reason to stay.

The business goals were just as clear: reduce dependency on third-party tools, remove unnecessary steps from the workflow, and drive higher activation and long-term use.

The redesign moved Graphics into Studio, as a tab in the same library users already browse for clips. Every dead end in the old flow became a step users can come back to.

Graphics now sits next to Library and Play-by-Play, with the same layout users already know: assets on the left, the canvas in the middle. Templates can be filtered by format, ratio and category, and Text, Stickers, Shapes, My Assets and Upload are one click away, the features users used to leave for.

As part of the process, we realized the templates themselves had to change. They were built in After Effects, which takes a long time to render and gives no immediate response, so there was no way to see an edit as you make it.

We moved the templates to Lottie, which gave us instant feedback and a fast preview. That shift is what made live preview possible, and it led to the insight, and the ability, to build templates that work across multiple ratios. Even complex templates, like a two-player comparison, now run on Lottie, with widgets updated to match.

This insight came straight from the user interviews. The same graphic is almost always needed in several formats, and in the old flow each ratio was a separate graphic built from scratch.

Now users pick every ratio they need up front, edit once, and every ratio is edited automatically. It was one of the most innovative parts of the redesign, and a real kicker.

• A preview from the first click. Selecting a template shows a temporary preview right away, with a short prompt on what to fill in next.
• Only what's needed. The side panel asks for a few user inputs, like player, game and main image. Most other parameters fill in from WSC's data.
• Changes update live. No more guessing until the graphic renders.

After the launch, activation of Graphics grew from 12.3% to 34.1%, an increase of 21.8 percentage points. More users stayed in Studio instead of switching to external tools, and content moved from creation to distribution faster, with fewer manual steps.

Graphics was close to being shut down. Its turnaround started with understanding the problem, not with redesigning screens.

StreamElements connects brands with thousands of streamers. Every sponsorship touches a brand, a product, a budget, an account manager and a list of creators, each with their own price, goals and dates.

All of that ran across three outdated systems that needed heavy manual work. Teams couldn't see spend or performance clearly, campaigns were slow to launch, and the tools held back growth.

Before drawing a single screen, I mapped every existing flow end to end, to see where the work actually broke down.

The first map defines the building blocks and how they relate: brands, managers, creators, deliverables (a stream or a social post), follow-up actions and notes. The second walks a real example, Blizzard promoting Diablo, through every screen: brand, offering, campaign, and groups that separate creators' pricing and flight times into buckets.

Wherever a flow needs an object, the map asks the same question: pick an existing one, or create it right there, without leaving the flow.

The final system grew to around 80 screens. Each section in the side menu, from sponsorships to campaigns and brands, has its own flow. The screens below are a selection that tells the core story.

The first version gave every object its own home. Brands, each with its industry and offerings, and campaigns, opening with the one number the team checks first: how many are running right now.

Below it, every campaign with its brand, status, budget, dates and account manager. Finished campaigns and brands move to an archive instead of crowding the list.

As the system grew, so did the navigation: Authorization, CRM, and Sponsorships, split into brands, campaigns, deployments and negotiations.

The campaigns table grew with it into a financial view. Next to each campaign sit its offering, budget, total spend and result, so the team sees where money comes from, where it goes, and what it brought back, without opening a report.

• Filters by what matters: date, creators, finances and status.
• Bulk actions: select campaigns to jump straight to their offers or creators.

Creating a campaign follows the model step by step. The first step asks only for the essentials: brand, offering, campaign type, account manager, and links to the website and Salesforce.

Each step builds on the last, so nobody has to hold the whole campaign in their head at once.

Choosing creators was the heart of the work. Each group keeps its dates, budget, conversion event and target CPE pinned at the top, so every choice is made against the goal.

Instead of browsing thousands of streamers, the team describes who they need, by location, stream and content, sets a maximum, and gets a list to review. The empty state says exactly how to start.

Creators are long-term partners, not line items. Each row brings together who they are, who manages them on our side and theirs, their recent campaigns and revenue, every channel with its numbers, and which StreamElements tools they already use.

With one system in place of three, setup and tracking were automated, and the team saw spend and performance as campaigns ran. Brands trusted the numbers, and budgets grew. Collaboration between StreamElements, brands and streamers became smoother, and campaigns smarter.