Interfaces where people and agents work together.
UX and interface design for systems where AI does part of the work: showing what the agent did, why, and what the person needs to decide.
Trust is designed, not announced.
When an AI prepares a decision, the interface determines whether people trust it, check it properly, or ignore it.
Good AI interfaces show the evidence, make the uncertain parts obvious and make it easy to correct the AI. Poor ones hide the reasoning, or bury the person in output they cannot review.
We design these interactions with the people who will use them, and test them on real cases before they ship.
What we build and run for you.
Agent workspaces
Run views, approval queues and evidence panels for people supervising agents.
Copilot interactions
Inline suggestions, citations and edit flows that fit existing screens.
Design systems
Component libraries with AI patterns built in, consistent across products.
Research and testing
User research, prototype testing and accessibility audits on real tasks.
Applications people and agents work in together.
Designed around the real workflow, engineered for the systems it has to reach, and supported by the team that built it.
Designed with users, tested on real cases.
Observe
researchWatch how people do the work today and where they hesitate.
Prototype
AI-backedInteractive prototypes that use real model output, not lorem ipsum.
Test
usersSessions on real cases to check people understand and trust what they see.
Systemise
libraryPatterns captured in a design system engineers can build from.
- Evidence visible for every AI suggestion
- Uncertainty shown, not hidden
- Easy override and correction
- WCAG 2.2 AA accessibility
- Bilingual and right-to-left layouts where needed
Where it earns its keep.
Illustrative patterns from the sectors we work in. Client details stay anonymised.
Officer approval queue
Decisions prepared by an agent, each with its evidence and a one-click route to the source documents.
Claims handler workspace
A single screen combining the claim file, the agent's assessment and the actions available.
Design system for AI products
Shared patterns for citations, confidence and agent status across a product family.
Guided applications
Forms that adapt to the applicant and explain requirements in plain language.
Chosen for the job, not the vendor.
Work this builds on.
A governed digital platform and partner API layer for a national education authority
A regulator's entire licensing lifecycle, rebuilt as one digital platform
Client details anonymised per delivery agreements.
Questions we get asked.
How do you handle accessibility and Core Web Vitals?
WCAG 2.1 AA is the default target for citizen-facing and government applications; Core Web Vitals budgets (LCP, CLS, INP) are enforced in the build pipeline, not measured post-launch. Both are designed in from the first sprint. Retrofitting accessibility at the end of a programme is more expensive than designing for it from the start; the industry has known this for a long time and continues to ignore it.
Do you design for Arabic and right-to-left?
Yes. Bilingual and right-to-left layouts are standard in our Gulf-region work, and we design and test both directions from the start rather than mirroring at the end.
Can you work with our existing design system?
Yes. We extend existing systems with the patterns AI interfaces need (citations, confidence, agent status, approval flows) rather than replacing what your teams already use.
What would you hand to an agent first?
Bring one workflow. In 60 minutes an AI architect maps where an agent helps, what it needs to reach, and what a first working version would take.