Quite Contrary: Kacper Staniul On Why Chat Is “The Worst Interface AI Has Ever Had”

“Chat is the worst interface AI has ever had, and we’re stuck with it.”

 

We’re living in a world of supposed binaries: you’re liberal, or you’re conservative; rich or poor; prejudiced or “tolerant”; hot or not (a nod to Paris Hilton and the millennials among us); and so the list goes on.

Of course, this isn’t the real world. Or rather, it doesn’t have to be. Reality is painted in different shades of grey, and things are only black and white if we choose to perceive them that way. Real life is full of nuance and opinions and ideas that you maybe half or only kind of agree with, or perhaps only in very specific contexts. But we seem to have reached a point in society in which we’ve become unwilling to navigate the complex.

Picking a side, saying “yay” or “nay”, is far easier than admitting that you actually fall somewhere in the middle, because that makes it messy, confusing and difficult to organise. Further to that, another subconscious reason we’re becoming increasingly inclined to go one way or the other, in my opinion, is that admitting you’re in the middle means that you don’t automatically get a team to back you up and support your thoughts and theories. And it does seem like that’s something people are reaching for these days. It’s almost like the best possible outcome within a discussion or even a disagreement is to join the side with the most support, so that you can be the loudest and hardest to ignore. Also pronounced, “the winner”.

But it doesn’t have to be this way.

In fact, I implore you – it doesn’t have to be this way. The only way we change this, however, is to change our own perceptions and our own behaviour. That is, find comfort in the unknown and the messy, and don’t be afraid to have an obscure opinion that doesn’t fall in one camp or another. Dare to speak your mind, to disagree or to partially, kind of, maybe agree just a little bit. And if you completely agree, then that’s just as important to talk about.

And in turn, we need to make other people comfortable doing the same. Don’t shut down opinions because they’re different to your own, because they’re confusing or because trying to understand somebody else’s perspective might be time-consuming or involve a little bit of friction. Ask questions, disagree and aim to understand – not agree, just understand, as far as you possibly can.

Because the more we run away from conflicting views and complicated opinions, the more we create false dilemmas and unnecesary black and white scenarios, the more we reinforce the “us versus them” mentality. And I’m sure we can all agree (is that possible?) that that’s not something we need any more of at this point in time.

So, let’s chat, let’s discuss. Let’s disagree! Let’s think, let’s ponder and let’s try to understand.

That’s the idea behind Quite Contrary, TechRound’s new series dedicated to hot takes, unpopular opinions and perspectives that might make people shift uncomfortably in their seats. Every edition starts with one deliberately contrarian viewpoint, and then we open the floor to the wider TechRound community.

 

 

This Week’s Contrarian

 

This week, our “unpopular opinion” comes from Kacper Staniul, Co-Founder of MyArchitectAI. Staniul believes that OpenAI’s ChatGPT has “the worst interface AI has ever had,” and not only that, but according to him, we don’t have much hope of getting away from it.

No doubt, this one’s sure to be controversial. Some people love Chat, others hate it. And most importantly, people seem to have plenty of different reasons why they feel the way that they do.

 

Kacper Staniul, Co-Founder of MyArchitechtAI

 

 

Kacper Staniul is the co-founder of MyArchitectAI, an AI-powered architectural rendering platform that helps architects and designers create photorealistic visualisations in seconds. A startup founder and growth specialist based in Estonia, he has built multiple internet products and is passionate about using AI to make complex creative workflows faster and more accessible.

 

Kacper’s Hot Take

 

“Chat is the worst interface AI has ever had, and we’re stuck with it.” 

 

“The blank text box is a design failure we’ve somehow decided to celebrate. All the work lands on the user: you have to know what to ask, how to phrase it, what context to paste in, what to try when the output misses. We spent forty years sanding friction out of interfaces — buttons, defaults, undo — and then launched the biggest capability jump in decades as an empty rectangle with a blinking cursor.

“Watch someone who isn’t a power user open ChatGPT. They type something short, get something generic, shrug, and close the tab. Usage plateaus at summarize-this and draft-that, and it’s not a model problem. The interface gives you nothing to hold onto: no menu of what’s possible, no hover state that teaches you anything. You either show up already knowing what to ask for, or you never find out.

“The AI products that win over the next decade won’t feel like AI products at all. The model gets buried in the workflow. And “”chatting with an AI”” ends up looking like the command line era of this whole thing: powerful, beloved by experts, and a phase.”

 

Here’s What the TechRound Community Had To Say

 

• Vladimir Beskorovainyi: Enterprise AI Architect and CTO,  Founder of Besk Tech (besk.tech)
• Debra Maxwell: CEO of ArvatoConnect
• Rav Hayer: Managing Director for UK/I and Head of BSFI Europe at Thoughtworks
• Justin Megawarne: CEO of Megaslice
• Borja Gonzalez Leon: Chief Scientist and Co-Founder of Iconic
• Inna Weiner: VP Product at AppsFlyer
• Luka Peric: Principal Designer at Infinum

 

Vladimir Beskorovainyi, Enterprise AI Architect and CTO,  Founder of Besk Tech (besk.tech)

 

 

“Chat is not a design failure. It is the only interface so far that is honest about uncertainty. Buttons and defaults work because the outcome is deterministic, and you cannot put a button on something that might refuse, hedge or invent. The blank box is what you get when the capability boundary is unknown even to the people who shipped it. Where I agree is that the model belongs buried in the workflow, and almost everything I build is exactly that. Nobody types into a box to classify a citizen request or query five hundred data marts. But the reason those systems work is not better UI. It is that someone narrowed the task until the boundary could be drawn. Chat will persist wherever nobody has done that work yet, which is most places.”

If chat is such a great interface, why does every AI company publish prompt guides?

“Because the guides teach the capability boundary, not the interface. Every general-purpose tool ships with a manual. Nobody calls the spreadsheet a failed interface because Excel courses exist. The prompt guide is a symptom of generality, and generality is the thing people actually wanted.”

 

Debra Maxwell, CEO of ArvatoConnect  

 

 

“I think we’re being a bit too quick to blame the chat box. The core issue is that most of us are still learning how to work as effectively as possible with AI. Too often, people type a quick one-line prompt and expect a brilliant, accurate answer, when the reality is that AI is only as good as the context it’s given.

“Treat AI just as you would a new colleague joining. The best results come when you brief them properly, with rich context: internal data, customer insight, and a clear objective. Using this as a guide helps colleagues ask better questions and connect AI to the knowledge that already sits across the business.

“Will chat be the dominant interface forever? Probably not. But whatever comes next, success will still come down to context. Just as your colleagues often need multiple drafts before reaching a final product, you need to make sure you’re giving AI feedback to polish, improve, and course-correct to get the best out of it.”

 

Rav Hayer is Managing Director for UK/I and Head of BSFI Europe at Thoughtworks 

 

 

“The rise of prompt guides is a tacit admission that we have shifted the cognitive burden from the technology back onto the user. At Thoughtworks, we believe technology should adapt to people rather than the other way around. When software requires a manual to explain how to ask it for help, the user experience has fundamentally failed.

“Great design creates empathy between people and systems. It uses context, visual cues and immediate feedback to make the next step obvious. A blank text box does none of that. It offers no guidance, no understanding of intent and no indication of what is possible. Instead, users are left guessing the right prompts and hoping for a useful result.

“Prompt guides are often presented as a feature, but they are really a workaround. They expose a design gap by asking humans to learn the language of the machine rather than building systems that understand the language of humans. The hard work of good design should never be outsourced to the user; it belongs embedded directly within the product itself.”

 

Justin Megawarne, CEO of Megaslice

 

 

“Every usable interface is actually an imposition. It’s an imposition on your freedom, your ability, your thinking and your goals. But we live with it for many things, because we just want to get something done. And that “something” is usually a very common, well-known thing. As soon as you want to do more than the interface allows, you are stuck.

“A blank chat textbox imposes nothing. That is the entire point. It’s freedom. Of course, the price is then that you have to know what you want, and you have to say it clearly. If you aren’t prepared to take that on, why is that the tool’s fault?

“Are we going to complain that code has a bad UI because Zapier exists?”

If chat is such a great interface, why does every AI company publish prompt guides?

“First of all, prompting is becoming less important as the AIs get better at figuring out your intent from their training and from asking for clarifications.

“Secondly, the chat box can accept almost anything. A lift button doesn’t need a manual because all it can do is go up or down. But every other general-purpose tool needs to be learned: Excel, cameras… or, you know, the English language.

“If you look at what those ‘prompting guides’ teach you, it’s basically good comms skills. Give context, say what you want, say what good looks like. That looks a hell of a lot like briefing someone well, doesn’t it?

“Sounds to me like we’re complaining about our inability to explain things well.”

 

Borja Gonzalez Leon, Chief Scientist and Co-Founder of Iconic 

 

 

“The problem with AI is not the method of prompting. It’s the way the models are designed and the role they play.

“Most LLMs behave like assistants, responding to our requests, solving problems, and providing answers, often within a single interaction. If that is the goal, open-format text entry is a good format for prompting.

“But if we want AI to do more for us and behave in a genuinely intelligent way, we need to move beyond the assistant model. That means developing products with a richer understanding of context, intent and objectives, rather than relying on the user to continually translate those objectives into prompts.

“Once that foundation is in place, different ways of interacting with AI, including diverse modalities such as voice, become more useful.

“The bigger opportunity is to make AI better at understanding people, rather than expecting people to become better at prompting AI.”

 

Inna Weiner, VP Product at AppsFlyer

 

 

“The blank box gets blamed for a problem that’s actually architectural. Most foundational LLMs already remember you within their own walls. That part is solved. The gap is between tools. Every new AI product still starts blind, so users end up re-explaining who they are and what they’re working on each time they switch context, and the real horror of this AI wave is uploading and downloading static files just to hand one tool what another already knows.

“Give context a way to travel with the user instead of living locked inside a single product, and the same “blank rectangle” turns into a fast, low-friction interface, because there’s no setup tax before every message. Conversation is one of the most natural interfaces humans have. What’s missing is portability, not structure.”

Luka Peric, Principal Designer at Infinum 

 

 

“I disagree. Chat is, in many ways, the most flexible interface we’ve ever built. You can type, paste a link, drop in an image or talk out loud and the model can work out what to do. Crucially, it is the only interface that can explain itself since you can also ask it what it can do.

“We’ve had this argument before. The search box was a blank rectangle with a blinking cursor, and it was criticised for putting the burden on the user. Yahoo’s directory offered a browsable map of everything you might want to find. It lost because telling a tool what you want and getting there directly is more seamless than navigating someone else’s categories.

“The real UX problem is what surrounds the chat. There are increasingly complex sidebars, tabs, projects, plugins and features scattered across the interface. That creates friction, particularly for newcomers. It is an argument for designing the product better, not abandoning chat as the interface.”

If chat is such a great interface, why does every AI company publish prompt guides teaching people how to use it?

“Prompt guides exist because AI is extraordinarily open-ended, not because chat is a bad interface. When the range of possible tasks is almost unlimited, there will inevitably be guidance on how to get the most from it.

“The nature of those guides has already changed. Early 2023 advice was largely about how to talk to the model, using magic phrases, elaborate prompts and “act as” instructions. Much of that has become easier as models have got better at understanding intent.

“What we’re learning now is less about phrasing prompts and more about using AI effectively: which model suits which task, when to provide context and how to structure complex work. That’s not evidence chat has failed. It’s evidence we’re learning to use a new tool. You wouldn’t use a flat-head screwdriver on a Phillips screw and knowing which to reach for isn’t a complaint about screwdrivers themselves. That’s a guide to the work, not to using the interface.

“Learning by doing is one approach, but this is new technology, so there’s a lot of knowledge sharing within the community. I think that’s beautiful.”

 

Want to share your hot take with the world? Contact us today