Your move

Croatia

Split
Dračevac 3D

Zagreb
Radnička cesta 39

Canada

Toronto
30 Commercial Road

Croatia

Split
Dračevac 3D

Zagreb
Radnička cesta 39

Canada

Toronto
30 Commercial Road

Testing Claude Design and ChatGPT Images 2.0: Why you still have to figure out what to build

Apr 28, 2026

Design

Reading time

3

mins

Author

Alessia Amanzi

It seems like people are still treating AI tools as if they are waiting for a “magic prompt” moment.

You type what you want, hit enter, and the tool perfectly resolves the entire project. When it doesn't, it's easy to dismiss it. 

On the flip side, there is a crowd propagating that Figma and traditional design are dead.

The reality of using these tools is much less dramatic. They are just another set of tools, and their value right now isn't in delivering perfection.

Let’s break this down. I will explain how these platforms fit into a daily workflow based on my own testing, followed by where the real utility lies.

For the last eight months, I have used Claude Code almost constantly.

My workflow lives in the terminal. I use it to build prototypes, test ideas, run brainstorming sessions, and build my own test apps for fun.

I am highly accustomed to this specific way of working.

The core benefit of Claude Code, Figma Make, and similar tools is that they allow me to take an abstract idea and immediately turn it into something tangible.

I can interact with it, test it, and evaluate it in a way that is simply impossible on a static screen.

With that context established, I recently tested the new Claude Design interface to see how it compares.

My immediate takeaway is that it doesn't offer anything I personally need.

If you already have a code-based workflow, the application is exactly the same, it is just wrapped in a different interface. 

For example, Claude Design markets "upload your design system" as a feature. 

I can already do that through the terminal by having Claude Code pull the components directly from the project repository.

Claude Design is just a layer of abstraction, not a layer of added flexibility.

The cost of the interface

That abstraction comes with a heavy tax.

Visual UI tools burn through usage limits significantly faster because they require massive amounts of background context and presets to function.

A single prompt in Claude Design ate through roughly 75% of my Pro plan limit

In contrast, working directly in the terminal with Claude Code allows me to comfortably work for almost a full eight hours on maximum settings.

This doesn't mean Claude Design is useless.

The benefit is a much lower barrier to entry. It provides a visual, less intimidating approach for non-developers, complete with shortcuts and presets.

It doesn’t bring any groundbreaking value to someone who is already comfortable in the code, but it opens this specific way of working to people who would never otherwise go in that direction.

The function of design

You can't argue that AI is exceptional at generating output.

But, you have to remember that design is the process of framing the problem and deciding what needs to exist in the first place, and tools like Claude Design won't do that for you.

Claude Design simply opens that workflow to a broader group of people (for example developers, marketers, founders, etc.).

It drastically shrinks the timeline from having an idea to holding a testable prototype.

I experienced this exact same shift with image generation.

Until recently, I used to ignore visual AI tools because they couldn't deliver the quality I needed for my work. That was until last week when everyone started talking about how Chat GPT Images 2.0 is a game changer.

Naturally, I started testing it, but mostly to mess around, running prompts to see if I could make my dog Greta fat.

After that I pushed it further and generated logo mockups for a project. The result was a 10/10.

The progress is immediately obvious. The model understands context, it is good with reading text, and it stays consistent across multiple iterations.

With Chat GPT Images 2.0 you can build pretty decent mockups on the fly for whatever concept you want in any style.

It is also surprisingly smart when you need it to adjust one specific detail.

Yes, it still makes mistakes. The model hallucinates, and sometimes you have to abandon the thread and jump into a new chat to reset its brain.

But it cuts out the manual grunt work entirely.

Now, instead of spending four hours digging through the internet looking for brand inspiration to see if the idea works, I spend five minutes.

As I said before, the point of these tools isn't to perfectly execute a final product. The value lies in skipping the tedious setup so you can immediately see if your idea holds up.Does that mean designers are in trouble? Far from it because AI technology will give us speed while we will be focused more on the clarity, the story, the clients needs. The people who will thrive in this environment are the ones who will bring taste and a strong point of view.

Alessia Amanzi, Design Team Lead at Profico