If you’ve done any sort of design recently, you most likely have used AI. Or, at least you’ve seen UI in your design tool of choice whether that be Canva, Figma, or even Loveable.

Even your everyday LLMs now support vibe design, as it were; including Google’s Gemini, and Claude which I have used for said purpose.

How AI is integrated into product design process is an even more important question than the use of AI in design. Depending on your workplace context, requirements and so on, using AI or not, can break a number of things.

After all some companies are declaring themselves AI first, while others require designers use AI. If the nuances of what exactly these means in practice are not hashed out, the outcome can be anything but.

UI Design with AI

On the surface, for quick prototyping on less complex use cases, AI does it. After that, it can be a miss or a hit depending on the complexity of your design cases.

Note I cannot speak to the code that is generated in such vibe coding, since I have no expertise whatsoever in that.

However, from what I have seen, the output of vibe design are simply weird, at least for now. For instance, I have tried to design a simple fintech app prototype on Gemini (Stitch) and the moment I mentioned WeChat, the text on the prototype was Chinese, which I do not speak.

As your design cases become complex or refined, the output starts to lose its design touch. For instance, color schemes get mixed up, even though now some tools feature options to set a design system. Text misalignments are another hiccup that AI just can’t get right, button naming and alignments, and so on, just to mention a few.

In other words, even after you’ve done your vibe design, you will have to do a lot of manual tweaking and adjustments, and review entire UI boards to make sure everything makes design sense.

AI designs are not original

But when you’re looking to do original design, intentionally from the ground up, current generation of vibe design tools will not do it for you. This is simply because they are trained and can only reference existing designs. As far as design goes, this is their biggest limitation.

For someone experimenting with design, doing quick prototypes, even looking to just copy some existing designs, I think AI can help, the issue of token costs aside.

If we are just moving fast and copying existing designs, then we risk enshittifying our products from the outset. A little later, we might find that designs have become homogeneous, bland and without taste.

Waiting on AI is weird

Another caveat of designing with AI is that you need to wait for AI to process your requests. This one I simply don’t get, especially when it takes long for the AI to complete a design task.

One, since you are never sure how long it will take to complete a task, you are not sure if to jump on something else while AI does its thing or to just wait. Waiting for AI or anything else for that matter is something we are not primed for, since doom scrolling became a thing.

Waiting on AI breaks the design workflow in at least a couple of ways.

Firstly, if you switch to another task, when you come back, you will have lost the context of what you were doing, and will have to retrace your steps. That is, assuming the AI already completed what you had asked and did not simply give up processing it.

Secondly, if you wait on AI in situ, it feels even more weird because it is not part of what you would call a conventional task workflow. Basically feels like you’re wasting time, and as if your brain has lost its capacity to do anything, because it cannot proceed to doing something else when the current subtask is being done on its behalf.

Certainly vibe design tools have thought about that, and have added status texts to the text-prompts input box, which you can read whilst you wait. So now you end up trying to read text that you simply can’t keep up with because it is moving too fast, and in any case have no need to read because it is simply telling you that AI is working on your request.

Thirdly, for me, this waiting on AI breaks the process through which designers develop tacit knowledge. See when you design, you realise and figure out a lot of small details why some information architecture will not work for your use case, and in the process your design rationales emerge. Now, when you wait, you’re not doing any of these, like I said, your brain is like in a waiting room and is not working.

AI will not end product design

This is the million dollar question and at the moment we can only speculate.

One thing to note for sure is that companies that care about design are creating a balance in how AI is used in product design, empowering designers to leverage the strengths of AI, while applying their skills to the process. Definitely there is more to design than just mere design boards, there is a design process, a product strategy, a monetisation strategy, and a lot of nuance and detail in user research, would you be willing to had over all that to ChatGPT?

The cost of token is obviously the elephant in the room, and with the impending possibility of token cost skyrocketing, I can imagine some companies will prefer to keep human designers as the key design resource.

Lastly but not least, companies, large and small, keen on creating original, carefully thought and designed products will most likely stick to their human designers. As I mentioned above, the skill of developing tacit design knowledge in a given domain over time is not easily automate able. In fact, with AI design, it gets lost in the process.

Privacy concerns is one last thing we can add here. Firstly, if you’re working on a sensitive, market leading, or some proprietary product, you do not want to feed your prompts to AI where they will be used to further train AI, or become outputs in the prompts of other users. In such cases, you might want to keep your human designers.

Your thoughts

Have you designed with AI? What has your experience been?