Product
We all built the same damn UX: a chat

Javier Bonilla
Co-founder of Querio
Apr 25, 2025
TL;DR;
As excitement around AI surged, everyone rushed to build chat-based interfaces—but at Querio, we learned that users wanted more than just text in/text out. Inspired by Mariia’s “Maybe Not?” blog, this post explores how we’re rebalancing the pendulum: keeping the magic of LLMs while bringing back structure, clarity, and good UX. Part one of a two-part series.
Our founding designer, Mariia, wrote a great blog post recently called Maybe not? She wrote a set of things we did to un-AI Querio and it serves as the inspiration for much of what I am about to write. There’s going to be a lot to say, so this blog is part one of a two part series.

The swing of the pendulum
There’s this thing I call the “swing of the pendulum”. It’s not trademarked or anything, it’s just my way of explaining how humanity tends to go very far in one direction when we get excited about something. And in the process of going so far, we forget about the lessons that got us there in the first place. Eventually, after hitting some wall, we realize we need to course-correct.
You see this pendulum everywhere once you start looking for it. Think about it: we get collectively obsessed with a new way of doing things, sprint full speed ahead, and then swing back toward a more balanced reality. Not all the way back to where we started, but somewhere in the middle where the exciting new thing and the proven old ways can play nicely together.
What I find fascinating isn't just that the pendulum swings, but how predictable the pattern is and yet how surprised we always seem when it happens. Every industry, every tech wave, same story. I'll give you some examples, and you'll see where I'm heading and why it's so related to what Mariia wrote about un-AI-ing our product.
The dot-com pendulum led Marc Cuban to become a billionaire, then the fundamentals of economics came crashing back, reminding us that companies actually need revenue. Twenty years later, the zirpies gave us another pendulum swing of "growth at all cost" unicorns, until the 2022 correction reminded us yet again that unit economics matter.
For us football nerds, Spain's 2008-2012 tiki-taka pendulum swung so far that by 2014-2018, Spain was just passing the ball like a headless chicken with no purpose. Now we've found balance again with a game where Lamine’s flair can co-exist with Rodri's control.
What about the "app for everything” pendulum which gave birth to the Zippo app. I, like many, briefly rocked it on my iPod, but eventually we realized that maybe not everything was going to become a mobile app and that real Zippos are still the best way to light a J.

Each time, the pendulum swings too far before finding equilibrium. Which brings me to AI or better put, LLMs.

The LLM pendulum
Early days
I don’t know about y’all, but it felt like ChatGPT came out of nowhere. Well, kind of. Looking back, I notice that the pendulum was gradually starting to swing, but I didn’t fully realize when I first saw it. I got to experience the emptiness of the beginning, the wild frenzy of the swing, and play my part on the way back to equilibrium. For once, I’ve had a seat—albeit a nose-bleed one—to the pendulum's entire arc.
Back in 2018 while studying at UT Austin, my friend (now co-founder) Rami had early access to IBM Watson’s natural language AI as he was building Lina (I’ll let him blog about that later). I had no idea what this was, to be honest, but he seemed pretty excited, so we spent some time playing with it.
This was our college days, so we loved playing the guitar and using real Zippos, but as non-English majors we were quite shit with words. So it made perfect sense to get this novel thing to fill-in the gap and write lyrics for us. We made a barebones app that would take in some songs you wanted as inspiration, fetch the lyrics and pass them as context, along with a description of what you wanted the song to be about. Out came the lyrics.
It was magical, but we had very limited credits, it frankly wasn't very good, and it died a slow death as a random cool toy we forgot about. Little did I know this was just my first sighting of a gentle push of a pendulum that would soon swing with much greater force.

ChatGPT wrapping
When ChatGPT launched, it instantly took me back to Rami’s college room. I was heavily impressed by how much better it had gotten during those years. Having already felt a similar magic before, it got me super excited about the possibilities. It turns out, this time it wasn’t just me who got excited—everyone was. And with that collective excitement, the pendulum had gained enough restoring force to blow past the equilibrium point.
Every product roadmap suddenly included an LLM feature and the pressure got so massive that even Apple, king sloth, announced AI so early that they’re now getting sued for it.
Now, there was a lot of talk about ChatGPT wrappers during that time, but they referred to the infra powering the engineering behind a product. Just passing prompts to OpenAI's API and returning results. As much as the critics want to disagree, there was genuine innovation behind the scenes wrapping those LLM calls. Infra, orchestrations, context, pruning, memory, tools— all carefully engineered systems our team will shed light on in future blogs.
Let's be real though: everything is a wrapper of something else, that’s how humans build. Cars are just fancy wheel wrappers. Books are alphabet wrappers. Software is a wrapper of wrappers down until you reach a bit. If you zoom out far enough, all of our accomplishments as a civilization are just wrappers of wrappers of ideas. The "just a wrapper" critique wasn't the gotcha people thought it was—it's literally how progress works and the critics will critique.

UX
That’s not what I am here to talk to you about. I want to talk about something a bit more subtle: the user experience we built around that hidden infrastructure. The pendulum didn’t just swing toward AI—it swung specifically toward the same interface.
We all built the same damn UX: a chat. Text in, text out? It must look like WhatsApp and operate like a linear chat. This was the pendulum swinging us so fast, that we forgot all our design principles. Chats are great for certain tasks— communication between humans as well as exploratory conversations and broad assistance with LLMs. But that was evidently not the right UX for the vast majority of applications of LLMs.
At Querio, we rode, or got carried by, this pendulum like everyone else. We built a chat. Then something interesting happened: our users started showing us through their behaviors that the pendulum needed to swing back. They wanted the AI pizzazz, absolutely—but they also wanted structure, guidance, and to be able to click around to change something instead of having to text their way to success. You know, basic UX design.
This is where Mariia's un-AI-ing of a product comes into play. We realized we needed to find that middle ground where AI's capabilities enhanced rather than replaced the interfaces and experiences users already understood. We needed to swing the pendulum back toward balance.
In Part 2 of this series, I'll dive deeper into exactly what we did with one of our core products to find this balance—how we kept the magic while stripping away the chat that was getting in our users' way. The pendulum never stops swinging, but knowing where it is in its arc can give you a major advantage.
