Leadership & Craft
Leading design for care
How I lead a UX/design team, and drive standards across a regulated platform.
How the team is set up
From three designers in 2024 to a team of twelve, with Leads, a Principal track and a dedicated UX researcher.
I joined Nourish in September 2024 as one of three designers. Two years on, we're a team of twelve (with four to six more hires planned), including Leads, a Principal track for senior designers who want to keep growing without managing, and our first dedicated UX researcher.
Most of my energy has gone into building a structure people can grow into, rather than just adding headcount. Two of our senior roles were filled by promoting people already on the team. When there was room for one more senior hire, I brought in a Lead to support a newly promoted Lead rather than adding another Senior designer. I also designed a bigger restructure and tested it on paper - then shelved it, because it wasn't feasible yet and a team growing this quickly needed fewer moving parts.
The first case I made for growing the team was about delivery. With three of us across the platform, product teams were waiting on design, and that was slowing delivery down.
We're structured around ownership of named roadmap goals rather than fixed product territories. Each goal has an owner, and people can move across goals as the work needs them.
The technical layer of the design system sits with a Platform Team, covering Figma builds, tokens, components, prototyping, handoff, automation and AI tooling. Coaching and community live in our guilds (the Design Guild and the Accessibility Guild), and the Accessibility Guild is being relaunched with sessions open to the whole business, so accessibility comes up long before design review.
Operating rhythm
A few regular touchpoints, plus feedback that runs all week.
- A weekly UX/Design standup
- A weekly UX/Design leadership sync with our Leads
- A fortnightly roadmap check-in with the whole team
- Structured, scheduled crits, alongside async feedback during the week
Crits sit on the calendar for a reason - a team doing regulated, high-stakes work needs feedback they can rely on turning up.
I write epics tied to our roadmap goals, and owners break them down into stories with their teams.
How decisions get made
Grounded in research and in what Pulse has already settled, and I'll make the call when something's pressing or clear-cut.
We're an open, honest team; where banter runs freely, so does constructive feedback, so a split opinion on a design is rarely shied away from and usually turns into a lively discussion.
On the whole, design decisions are litigated against research, or against solutions that have already been tested and established, often in Pulse. If something's pressing or genuinely clear-cut, I'll step in and make the call myself.
Most of my time goes into keeping an eye across product areas and setting direction and standards, but I still get into the design weeds myself, including at canvas level. Understanding first-hand what the team is working with keeps evolving the lens I lead through.
Making the case for design
How design's seat at the table has grown alongside the wider business.
When the business set up its impact framework, I took ownership of the satisfaction measures - and kept them about how people actually feel using the product, rather than throughput or ROI. Throughput already had an owner, and nothing else in the data was giving frontline care workers a voice.
As AI-assisted prototyping sped up the route from design to delivery, there was little appetite for hard gates. So I proposed that steps can be skipped as long as every skip is logged, which keeps decisions visible without slowing anyone down, and design is now running the governance pilot.
Design now comments on design system component PRs alongside developers, and the tooling to make design approval a required step is already in place. Developers backed this from the start, and so did we; what it needed was real capacity and clear ownership, so it's now a rotating responsibility across our senior designers and above.
Not every case lands first time. Early in an internal AI pilot, I raised a regulatory risk with leadership as a warning, and it didn't get much traction - the energy in the room was (understandably) on what was being built, and my concern came across as a brake.
So I changed how I made the case. Rather than arguing the risk, I designed for it: the AI points people to the evidence instead of telling them what to conclude. I annotated the real screens to show where certainty was overstated and how the copy should change, and framed the work around protecting the launch rather than slowing it down. When it came up again, engineering backed the approach in the room.
The wider question still needs an owner, so I'd call it a partial win; we changed how the product behaves, and the governance is still a work in progress.
The standards design owns
The tenets and goals our team designs to.
We define these standards in partnership with Clinical Safety, engineering and PMs, and shape them through research and testing with users rather than in isolation.
Six tenets
Each tenet comes with a question we bring into crit.
- Inform, don't decide. Design informs decisions, and people make them. Automation and AI can suggest and explain, but they always defer to a person. Who makes the call here, and does the design make that clear?
- Prevent, then forgive. Design out the likely error first, then make every action recoverable. What's the most likely mistake, and how does the design stop or undo it?
- Design for the real day. Interruptions, shared devices, patchy wifi - and every state, including empty, loading, error and offline. What happens when this goes wrong, gets interrupted, or only half loads?
- Attention is earned. Every element earns its place by changing what someone does, with details one tap away. What does this make someone do differently?
- Words are care. Terms, labels and tone get the same care as layout, and they're agreed before anything gets built. Would the person this is about be comfortable reading it?
- Decide out loud. Show your evidence, label what's still a hypothesis, and record each decision with an owner - whether a person or AI built it first. What's the evidence, and who decided?
Three goals
This year our team owns three goals. With accessibility, we're moving from WCAG as the ceiling to a care-contextual standard, with WCAG 2.2 AA as the baseline. With AI workflows, we're moving from scattered individual exploration to a shared, documented way of working. And with Pulse Voice, we're setting one standard for how the platform speaks and shows up, in words and imagery - because how you word a safeguarding notification or an end-of-life journey affects people's dignity and safety.
Reviews and documentation
Design reviews design system component PRs in GitHub alongside developers, and our design system documentation lives in Supernova.
Designing with AI
AI absorbs the production work, whilst people keep the judgement calls.
We use AI to absorb production work like components, tokens and documentation. Judgement on experience, safety and trust stays with people, because what we ship reaches care professionals and the people they support.
The build loop
We're piloting a loop that takes a component from Figma to shipped code and back again.
- AI builds the Pulse component in code, with visuals coming from Figma through Figma MCP and rules from our Pulse skill files.
- Storybook and automated tests run.
- Design and development review the pull request together, which is where people sign off.
- It merges and the package is published.
- AI regenerates the Figma variants from the shipped code, so design and code stay in step without manual redlines.
Skill files
Our Pulse skill files are machine-readable rules (colour, type, accessibility, content language, components, inputs and navigation) that AI tools load before they start, so their output begins on-system rather than being corrected later.
Every designer rotates through AI-assisted builds, whilst system review and accessibility review stay with named, fixed owners.
Rollout
Rolling AI out to the team wasn't a mandate, but a series of design sprints and ad-hoc experiments - testing Claude and Figma MCP against real build problems, and discussing the output openly in crit like any other design decision.
Principles
These are our team's AI principles, published in our design system docs:
- Clarity over magic - AI output is a starting point for discussion, never a prescription
- Protect the problem space - AI assists discovery, it doesn't shortcut it
- Strengthen, not short-circuit - AI integrates into a human-centred process rather than replacing stages of it
- Prioritise learning - time saved goes back into research, testing and iteration
- Human at the helm - every AI-assisted output is assessed for bias, user alignment and strategic fit
Designing AI for care
Content standards for the AI features care workers read.
Once AI starts generating content a care professional reads directly, someone has to own what it can say and what it must never claim. At Nourish, that sits with design.
Our first tenet (inform, don't decide) matters most here. People tend to stop checking a system that sounds confident, and software that interprets clinical data and directs action can also count as a medical device under UK regulation.
So far our standards cover content:
- Disclosure language that makes machine-generated content clear
- A tone standard for that content
- Hard limits - never asserting clinical judgement, or speculating about care outcomes
- Predicted content kept clearly marked
- Copy that steers people to review the information, rather than accept a conclusion
We trialled the first version of these standards on an internal AI feature, and fed what we learned back into them.
What we don't have yet are interaction patterns for how people come to trust an AI feature, notice when it's wrong and recover from it - and that's the next piece of work.
Nurturing design talent
470 applications, around 100 interviews and more than 10 hires since late 2024.
Hiring here is close to constant, driven mostly by growth rather than one-off backfill. Since joining Nourish in late 2024, I've reviewed roughly 470 applications (380 design, 90 UX research), run around 90 full interviews plus another 10 lightning rounds, and made more than 10 permanent hires.
At that volume my process has to hold up under repetition, with the same structure and the same criteria, so that a yes means the same thing in January as it does the following July.
Bringing people in is only half of it; the other half is keeping them empowered and progressing once they're here. Alongside regular 1:1s and team touchpoints, I track capability against a multi-level competency framework built around what each role actually needs, which I've recently updated to cover the Principal track.