top of page

AI-ASSISTED DESIGN PRACTICE

The Uniformity Instinct: What AI Design Tools Get Wrong by Default

Alex Thanasenaris - 9 min read

AI Workflow

Design Judgement

Claude Code

Systems Thinking

Over the past few weeks I ran three small, unrelated experiments with Claude and Claude Code — an enterprise security dashboard, a personal tool for tracking my own design process, and an AI-native admin console. Different domains, different stakes, different scope. Each one broke in the same way on the first pass.

I want to walk through all three, because the repetition is the actual finding. One coincidence is a fluke. Three is a pattern worth naming.

Three unrelated projects, one shared bug

01

Secuirty dashboard

Four different severity scales, flattened into one generic High/Medium/Low badge — erasing the exact information the product existed to preserve.

03

AI-native console

A brief explicitly warning against "generic chatbot" still produced a chat window, with a vague, silently-executing confirmation.

02

Personal tracker

A full pasted prompt and a two-word screenshot note, given the identical one-line input field.

article-1-diagram-pattern (1).png

Why this keeps happening

 

This isn't a training failure. Uniformity is what "clean design" looks like by default, absent an explicit reason not to flatten. Nielsen Norman Group has framed the broader shift underway — from interfaces built around discrete commands to interfaces built around inferred intent — as the first genuinely new UI paradigm in six decades, and that same shift is exactly where this flattening instinct hides.

3/3

experiments hit the same failure

5/6

non-negotiable constraints missed on the console's first pass

What stayed constant across scale

 

The critique instinct required to catch this was identical whether the stakes were an enterprise compliance risk or a two-word screenshot note. That consistency is the real argument for why design leadership experience doesn't get less relevant as these tools get more capable — it gets tested more often, at a much faster cadence.

"What did this quietly treat as the same, that isn't?"

The reframe

 

AI didn't lower the bar for design judgment. It raised the frequency at which judgment gets called on. The acceleration is real — three working, critiqued, iterated prototypes in the time it used to take to finish wireframes for one. That moment of judgment didn't get automated. If anything, there are more of them now, not fewer.

SOURCES REFERENCED

 

Nielsen Norman Group —  on the shift from command-based to intent-based interaction

Industry writing on AI-generated polish and interface trust in 2026

© Copyright 2026 - Alex Thanasenaris - All rights reserved. 

bottom of page