24 September 2026 · Philipp John
Designing an AI product for a large organisation
Sources, roles, long tasks and transparency. What I learned designing an internal AI platform end to end, and what the EU AI Act changes for interfaces.
Design an AI product around trust. Show where answers come from, make roles and permissions visible, give long tasks a clear state, and tell people when they are talking to an AI. Since 2 August 2026 the last point is also required by the EU AI Act for many systems.
I designed an internal AI platform for a global insurance group end to end. This is what the work taught me.
What is different about an AI product?
A normal interface does what you click. An AI product answers in its own words, sometimes wrongly, and sometimes after a long wait. Users need to know what they can rely on, who can see what and what is still running. Those questions become the core of the design.
What did the benchmark show?
Before exploring directions, I benchmarked how leading AI tools handle four things: sources, projects, roles and long tasks. These became the structure of the product:
- Sources. Show what an answer is based on, and let people check it.
- Projects. Give work a place, so a conversation is not the only container.
- Roles. Different people need different starting points and different permissions.
- Long tasks. Show progress and let people leave and come back.
How do roles shape the interface?
The clickable prototype of the platform shows one product in four views. Switch the role, and the start cards, the signals and the locked features change. A claims specialist and a marketing manager use the same platform but see what is relevant to them. Locked features are shown as locked, not hidden, so people understand what exists and why they cannot use it yet.
How do you design for a regulated environment?
Answers have to follow the brand, CX and claims standards of the organisation, and strict data rules. The EU AI Act adds transparency duties. They have applied since 2 August 2026 and include:
- People must be told that they are interacting with an AI, unless that is obvious.
- AI-generated text, images, audio and video must be marked in a machine-readable way.
- Fines can reach 15 million euros or 3% of worldwide annual turnover, whichever is higher.
Generative systems that were on the market before 2 August 2026 have until 2 December 2026 to meet the marking duty. Check the details for your case with a specialist.
For design this means an AI label in the interface, a clear way to see sources and a visible state for generated content. Plan these as components in the design system, not as a late patch.
Why does the design system matter here?
An AI product needs components that regular systems do not have: a source list, a review and approval step, a confidence or status signal, a long-task progress view. On this project I extended the client’s design system with a dark mode and the components AI needs. Then the prototype could be built on the same system the product team would use. See how prototypes on a real design system work.
Key takeaways
- Build trust with sources, visible roles and clear task states.
- Show locked features, do not hide them.
- Treat AI transparency as a component, not a legal footnote.
- Extend the design system for AI patterns early.
The Insurance Inhouse App case study is protected with a password. Ask for access.