The attack of the AI design slop monster (a survivor’s guide)

In 2026, AI-assisted UX design has arrived. Claude, Codex and other mighty AI tools can now generate websites, design systems and working prototypes in minutes.

Or can they?

Many of us now recognise the signs of AI design slop: odd colours, bloated rounded corners, strange UI patterns and franken-interfaces that feel like a bad fever dream. Some of us have watched agentic AI overwrite files after an approval prompt no human could understand. Others have seen junior team members trust AI-generated designs without knowing why the output is weak. And many UX designers have come across the newly empowered non-designer who can now turn an idea into a prototype and take it straight to senior leadership before anyone has defined a user need, tested a hypothesis or opened Figma.

This talk is about how to push back and reclaim your UX designer groove.

I’ll share how I built a practical defence against AI design slop: a non-negotiable UX research backbone, clear user hypotheses, a skeleton UI structure for AI to work within, and enough controlled experimentation to move fast without letting the tool drive the work. I’ll also cover what happened when I produced some slop of my own, built a mostly non-sloppy prototype faster than a non-designer could vibe code their version, broken and rebuilt team relationships along the way, and finally got hold of the coveted MCP solution, the closest thing I’ve found to an antidote from the slop so far.

About the speaker:

A UX lead for Tolley+, a suite of digital and AI products for UK tax professionals, part of LexisNexis. I lead UX designers and researchers across multiple products and workstreams. After helping release the UK’s first AI solution for tax professionals over a year ago, I now focus on developing the next generation of advanced AI tools for our users. I use AI extensively in my own design and research practice. I graduated from City St George’s in 2021 after switching careers from architecture.