AI in government design: Is ‘vibe design’ ready for public sector projects?

Louis Pattison

AI design tools are everywhere but is “vibe design” ready for government UX? Our user-centred design lead, Louis Pattison, investigates the role of AI in government design.

The last couple of years have seen a rapid growth in the marketplace for AI-powered ‘vibe design’ tools. Apps like UX Pilot, Lovable, Cursor and Stitch – as well as new features built into established design tools, such as Figma Make – promise to dramatically accelerate the design process, suggesting that anyone can design pixel-perfect user interfaces in a matter of seconds with just a few carefully worded prompts. 

It’s an enticing idea. Design can be a tricky and time-consuming process, especially in government, where designers must be across everything — from the GOV.UK Service Standard and departmental design systems to accessibility requirements and the latest findings from user research. 

As consultants to the government, we’re ultimately paid for by the public purse, so it’s responsible to explore ways to speed up the process.

But as anyone who’s spent time with generative AI tools will know, they can be somewhat hit and miss – subject to hallucinations, overly literal, sometimes forgetful, and prone to not showing their workings. 

While we’re always looking for ways to improve our processes, when it comes to design, we – and our customers – have high standards.  

Exploring the potential

A lot of the experimentation and innovation work at Zaizi takes place within our communities of practice. A community of practice is a small group of people with a shared skillset or area of expertise. We meet regularly outside of the project environment to share knowledge and test out new ways of working. 

For the last few months, our user-centred design (UCD) community has been experimenting with these tools outside of our day-to-day project work to see what they can do. 

Are vibe design tools currently unfit for GOV.UK standards?

What some of these tools can do is impressive. Our experimentation showed that you can quickly build user journeys across multiple pages from a few written prompts. Tools like Lovable and Figma Make let you extract code from your designs, making them functional prototypes.

Figma also has AI-powered features that streamline some more onerous aspects of the design experience — for example quickly renaming layers in a design. 

But there are a few drawbacks to the current vibe design tools that, for an organisation like us, make them not fit for purpose. 

They’re inexact 

In July 2018, GDS launched the GOV.UK design system. A design system offers a single, consistent source of styles, components, and patterns. This brings with it huge benefits. It’s good for users – offering familiarity and supporting understanding through consistent approaches rooted in evidence. And it’s good for designers — speeding up design work and meaning they can focus on the real, knotty problems within a service. Not all the departments we work with adhere to the GOV.UK design system. But most use a design system of some sort, and it’s imperative that the designs we create follow them closely.

Unfortunately, while vibe design tools can ingest design systems, they don’t always have the context to apply them accurately or consistently. You end up with designs that look a bit like GOV.UK interfaces but a bit ‘off’ — the spacing is wrong, components inaccurately rendered, headlines and subheadlines are too big or too small. It’s possible to fix some of these problems with further prompting or even manual fixing — but the point stands that in this kind of work, precision is key. 

Which brings us to the next point…

Download our AI in government white paper — discover how to make AI simpler, safer, and stronger in your organisation

They don’t save time

One of the big promises of vibe design is that it’s a time-saver — type in a prompt and it quickly spins up a complete user journey. Tools like Figma Make and Lovable can do this, to an extent. The results can be useful at that early ideation stage — as long as you don’t mind treating the results as an experimental sketch, rather than something to build on. 

The idea this would bring big time-savings, though, is the result of a misconception about exactly what designers do. Building an initial user journey can be pretty quick, particularly when you’re working within a modular design system. The time-consuming aspects of design come after, when you’re iterating in response to user research insights or stakeholder feedback. 

This is real shifting pixels stuff — granular and precise — and not the kind of work that you would want to leave to a generalised prompt.

They may not be secure

Much of our work at Zaizi is in the field of national security. That means that we’re sometimes designing services that contain sensitive information. 

Naturally, this informs the kind of tooling that we use in our day to day work. If we’re putting data into a tool, we need to know that data is secure. But AI is your classic “black box” — a system whose internal workings are a mystery to its users. 

We can’t be sure that information that goes into an AI tool won’t find its way out again in some way. That’s why we have a policy not to expose sensitive project material to artificial intelligence — currently we don’t know enough about how these tools work to be sure where the data goes. 

At Zaizi, our policy is clear: sensitive material stays secure. 

Beyond vibe — uncovering where AI can help

In user-centred design, as in AI, things move quickly. Five years ago, the digital prototyping app Figma was the new kid on the block. Now, it’s the industry standard. In five years time it might have been swept away by something new. 

AI will likely be part of the next shift, so our UCD community is going to keep testing these sorts of tools to fully understand their capabilities, strengths and weaknesses. 

We’re also examining our internal workflows, looking for pain points in our current processes and working out where we can apply generative AI to make our work stronger and more efficient, while insisting on safety and security.

In our subsequent blogs in this ‘AI in government design’ series, we’ll explore the role user-centred design plays when integrating AI into your government services. We’ll show you why a ‘human-in-the-loop’ is non-negotiable and how to sharpen your user research with AI.

Subscribe for updates for our next blog in the series and to find out more about our upcoming events.

Related content

Thanks for joining us! We’ll keep you informed with regular updates.

Subscribe to our monthly newsletter, including updates about our content and events

Consent(Required)
This field is for validation purposes and should be left unchanged.