Product Designer vs GPT Astra 6. A real insight by a Product Designer

The Real Constraint Nobody Talks About

Hello everyone and today i am actually very excited to write this article manually after 2019 because Ai seems not to be understanding what I wanna say. Before I start let me tell you I am Tatheer a product designer wanted to break into health tech. I have over 4 years of experience and i was really upset because i was applying to roles and was not hearing back anything although the situation is still the same. I was so much stressed about the job and GPT just released Astra 6 model and hype was real I mean i was amazed by those videos and was asking myself if I am just replaced by ai like really this thing was hurting me out because back in 2018 I learned seo and soon the news spread on internet that seo is dead I was 14 or 15 years old by that time so i left it because i was not wise. Spoiler Alert Ai is just a fancy way of using template and i will prove it later.

However, it’s Big 2026 and people are still earning money as SEO expert and guess who is not obviously me and you lol. Now back to the thing that i was scared so i decided why not to compare myself side by side with GPT Astra. Honestly i was not afraid even a little but deep down something in me was questioning my skills. Theres only one thing echoing in my ears

“Leave the design before design leaves you“

Fast track to the comparison I never had GPT Ultra subscription before so i invested just to try the model and check the cpabilities.

Before I even talk about the test, here is the issue with how AI design tools are being sold to us right now. They are marketed as if the model understands design. It does not. It understands patterns and weighs everything in pixels. It has seen millions of landing pages, millions of dashboards, millions of Figma files, and it has learned what design usually looks like. That is a completely different skill from knowing what design should feel like for a specific brand, a specific user, or a specific business problem.

That is the constraint. Not intelligence. Taste. A model can be trained on every UI ever shipped and still have zero understanding of why a human eye rests longer on one layout than another, or why a five pixel shift in spacing changes how “expensive” a brand feels. That gap is exactly what I wanted to test, on myself, honestly.

Why I Picked These Two Specific Tests

I did not want to cheat. I did not pick a random Dribbble brief or something with no real stakes. I picked two projects that already existed in the real world before I ever touched AI.

The first was the Motionhash landing page, a project I had already designed myself. The second was a landing page challenge I got during an actual interview process with an agency a few weeks earlier, one I had already solved on my own too. Both had a “before” version already locked in. That meant there was no room for me to unconsciously bias the AI’s output toward looking better, and no room for me to cheat by mixing my own decisions into its process. It was a clean, honest comparison against work that already existed.

Test One: No Brief, No Mercy

I gave it nothing. No assets, no direction, just “design a landing page for Motionhash.” It went and pulled information straight from our live website and built a full page in Figma on its own.

Here is where it actually surprised me. The file itself was clean. Layers were named properly, sections were broken down logically, everything was organized the way a well trained junior would structure a file. That part is real, and I will get back to why that matters later.

But the actual design was just acceptable. Nothing more. It was more of copy of our original design but ignoring intentional choices and trying to fit everything without thinking how it will matter. It looked like a landing page template because it statistically resembled a thousand landing pages it had seen before, not because any real decision was made about how Motionhash should feel to a visitor. My own version was not even in the same league. Although Ai version has small Ux flaws but that could be easily fixed by observing .

Test Two: Full Brief, Real Stakes

This time I gave it everything. Assets, reference sites, the previous version of the site, and a detailed brief, exactly what I was given in the actual interview challenge take home task.

It struggled immediately. First attempt, it sat there thinking for over twenty minutes and still could not produce a landing page it was my nd generation and it kept thinking until limit exceeds. Second attempt, around thirty minutes, hit some kind of limit, and instead of a proper Figma file it gave me an HTML version. Acceptable, but template shaped again. Now I have tried asking for figma file multiple times but it was always making a html website.

Before i conclude anything I was using GPT Astra 6 at Max effort.

Although i could critique design myself but as a ux designer its my duty to take review from someone else and i conducted a short research where i asked 5 people about which one is better now here hear me it is becoming more of UI comparison rather then a UX one. Because on landing page UX part is not that much hard to cover and i will later explain it in detail where AI is failing in UX.

Now back to people all of them never new which one is ai and which one i designer, compare both designs side by side. Mine held up as the more premium one. It still needed refinement, because your own eye gets tired staring at your own work too long and outside opinions always sharpen it further and could hold the title of better one. But between the two, mine had the vibe. The AI’s did not.

Now here is the part I actually want to be clear about, because it is easy to misread all of this as “I beat the AI.” That is not the headline. The real headline is this is not me vs AI. It is me with AI. The tests were to find out exactly where the line sits between what it can carry and what it cannot, so I know precisely where to lean on it and where to trust myself. You may have heard the phrase that Ai wont replace you but someone using AI will, and its right at some poin now again i never said its fully right i have reasons for that too.

Now you may be afraid that how we will got the job i will clarify later in article that which is the real market of AI rather then just making you afraid that quit the design.

Where AI Actually Wins, and Why I Am Stealing It

The one thing I cannot deny, and honestly do not want to deny, is how well it structures and names layers. That is not a small thing. That is actual production hygiene most designers, including experienced ones, get lazy about under deadline pressure. I am genuinely going to start implementing that discipline into my own workflow, because there is no reason to be worse at organization than a machine that does not even understand what it is organizing.

Beyond that, here is where it earns its place:

• Generating individual components fast
• Building an initial design system to react to, not to ship
• Testing multiple directions quickly before committing
• Producing edge cases and variations faster than doing them one by one manually
• Giving a rough skeleton I can then tear apart and rebuild with intent

The Fake Speed Problem

Everyone keeps repeating that AI makes design faster. But heres me Tatheer the designer having a opinion against the market. In practice, for anyone who already has the skill, it often does the opposite. Prompting, waiting, reviewing a mediocre output, prompting again, fixing it manually anyway, that loop can genuinely take longer than just designing it yourself in the first place. It feels productive because you are always doing something, typing prompts, waiting on generations, but that is motion, not progress. The real speed only shows up on the specific tasks I listed above, not on full page or full flow generation. But I always fine it fast to make something new or a component in few clicks and it works.

The Truth About “Agentic” AI

You have heard on internet that AI is simulating world this might be true but believe me Agentic is still a word now i am not ai expert and i have my own opinion if you are an ai expert then i really love to learn more about it leave a comment because i am always eager to learn new things. Back to our topic Everyone is throwing around the word agentic like it changes the fundamental issue. It does not, at least not yet. Agentic just means it can now chain more steps together and use more tools on its own without you prompting each step manually. That is a workflow improvement, not a taste improvement. A model that can now open Figma, pull brand assets, and generate a page in one flow instead of five separate prompts is still making the exact same kind of decision at every step, a pattern match. Chaining more pattern matches together does not create judgment. It just makes the lack of judgment move faster and touch more of the pipeline at once.

AI Is a Fancy Way of Using Templates, Here Is Why

We already had component libraries before any of this AI hype started. You can find many in figma community or by oficial brands for example material designs by google, UI kits, design systems shared across teams, all of it existed and was already speeding up production for anyone willing to use it. What AI does is essentially automate the act of picking and combining those same patterns based on a prompt instead of a manual search. That is useful. It is not new intelligence. Both landing pages it generated for me leaned heavily on the references and existing sites I gave it. It was remixing, not creating something that did not already exist in some form. Now it does not mean it can genertae new we can actually provide references moodboard and sketches to make our generation better and now you may have felt that it will actually take similar time in manual way and with AI.

If your actual argument is that AI generating a button or a card layout is going to replace designers, I am going to be straightforward with you, step out of the rock bro. We have had reusable components for years. Nobody lost a job to a component library. This is the same category of tool, just with a chat interface on top of it.

Where AI falls Apart the problem

Now if we give a complex problem to Ai then it will struggle alot in maintaining an effective UI with better UX. Assume that Ai has made the Ui then it still wont be able to generate the real experience, it as i said help to test ideas in bulk. Let me be clear if you are looking for something specific and complex in your fintech product or lets say medical product then AI wont be able to handle that and thats actually the way to break into this industry rather then simple landing pages.

Where the Market Actually Sits Today

A brand new founder pre funding usually does not care much about brand identity. They care about shipping something that proves the idea works. At that stage AI generated design is genuinely good enough, fast, cheap, and the bar is intentionally low because the priority is validation and MVP, not polish.

The moment that same founder raises real funding, seed or Series A, priorities shift almost overnight. Brand identity and design quality suddenly become a business priority instead of an afterthought. That is exactly where templated AI output stops being sufficient and a real designer becomes necessary again.

So here is where to actually break in if you are early in your career right now. Do not try to compete on landing pages and basic marketing sites, that tier is commoditized. Compete on systems thinking, on products with real state logic, real edge cases, and real user psychology. That is the layer AI genuinely struggles with, and that is the layer that still needs a human badly.

What To Learn, and What Not To Worry About

Learn how to think in systems, not just screens. Learn to read a user’s frustration before it happens instead of just decorating a flow after someone else designed the logic. Learn to use AI as a first draft generator for the boring 80 percent so you can spend your energy on the 20 percent that actually needs a human eye.

Do not worry about a model that can arrange components. That was never the hard part of this job. The hard part was always knowing which fifty decisions to make before you even open Figma, and that part has not moved an inch.

As a Product Designer, Here Is What I Actually Think

Honestly i am not the words best designer but i am not the weakest either, I might not the strongest in raw UI execution inside Figma but it will take me a day or two to improve and i will remember that, and I will say that plainly. But my taste is good, and I understand the human eye and the UX layer well enough to catch what AI cannot even see. That gap works both ways.I know what is necessary for a business. AI covers my execution weaknesses, my taste covers its complete lack of one. That is not me losing to AI or beating it, that is exactly why the combination is dangerous in a good way.

As a Designer, Here Is What I Think Your Responsibility Actually Is Now

Your job is no longer just to produce a screen. Anyone with a subscription can produce a screen now. Your responsibility is to be the filter, the person who knows when the output is template shaped garbage dressed up nicely, and the person who can tell a founder honestly that what the AI generated will not hold up once real users or real investors are looking at it. If you cannot tell the difference between a templated design and an intentional one, you are not being replaced by AI, you were already replaceable before AI showed up.

When I Will Actually Use It Going Forward

I am not writing this to dismiss it entirely, that would be dishonest after everything I just laid out about where it wins. Here is exactly where I will bring it into my own process from now on. Basically a good PD with AI is a lethal combo. I will feed it rough sketches and moodboards I already have a direction on, not blank prompts, so it has an actual creative anchor instead of guessing from scratch. From there I will let it generate components, quick variations, and edge cases off that foundation, and I will take over for anything that needs real judgment, brand feel, or a decision about how a user should feel in a specific moment.

Leave the Design Before the Design Leaves You

There is an old line famous in football for ronaldo btw he is the real goat, leave the game before the game leaves you, meaning get out on your own terms before you are forced out. I am borrowing that for design. I did not hold back in either of these tests. I did not pick easy briefs, I did not simplify the challenge, and I did not go easy on the comparison, because I genuinely wanted to know where I actually stand in today’s market before the market makes that decision for me. If AI was already better than me at real creative direction, I would rather find that out from an honest test on my own terms than get blindsided by it during an actual job hunt. However i am still looking for a job and if you know anyone who could help me to find one do connect with me on linkedin

Who This Is Actually For

If you are a solo founder or a small business trying to get something live fast without a real design budget, AI is genuinely useful for you right now, use it without guilt.

If you are a funded company, an agency, or anyone whose product depends on real brand equity and complex user flows, you still need a human who can think in systems and has taste. Nothing in these tests changed that.

And if you are a designer reading this wondering whether you should panic, the answer is no, but you should absolutely get better. The bar just moved up, it did not disappear.

Product Designer vs GPT Astra 6. A real insight by a Product Designer was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.