I designed an AI app. Here’s how it went.
The year was 2024. ChatGPT was more than a year old and it was taking over the internet, and Microsoft was going around selling their Copilot app as a corporate-friendly ChatGPT solution. A VP at the bank I work at had an idea: how about we make our own GPT wrapper app? We can tailor its features to our own needs and it might be cheaper. And so CIBC AI or CAI was born.
Conceptualizing
CAI was positioned as a work app from the beginning, and at the time, LLMs were known for being great for fetching information from text-based documents and synthesizing. The first few wireframes I made had a very clear focus on this use case – there was this assumption that users would often bring their own documents into a chat.

The Saved panel idea was dropped rather quickly in favour of simplicity (ChatGPT literally just had a chat bar at the time).
With a bit of polishing, things started to take shape:

Old thoughts
Admittedly, some UX considerations at the time were very “old school” when looking back at them. Please witness this very convoluted way of trying to make sure the LLM knows which file the user was referring to.


This was before we knew that you could just type the file name and the language model would be good enough to understand what you were getting at. Sigh.
Ambitions
All the major / popular AI apps were bringing out new features on a 24-hour interval. Under their influence, it was only fair that our team also had new feature ideas every single week. Some of them made it to production, some of them didn’t or ended up being added much later in a very different way. Here are some drafts for a few of them:


An admin dashboard and a search feature:
A moment in time
I dropped from the CAI team in mid 2025, shortly after its bank-wide launch. There were some other stuff that I really wanted to get back to, and a 50:50 time split wasn’t feasible for me due to how fast-pace CAI needed to be.
At the time of launching CAI to the entire bank, it was a chat app with various “recipes” (think custom GPTs in ChatGPT or Gems in Gemini) to help with specific tasks.


Since I left, CAI has gone through many more feature enhancements and a major redesign. It now has a marketplace, smart libraries, connectors to other apps, and looks nothing like what I’m showing here. I’m happy to be part of the journey and it was cool to see a product evolve based on rapidly-changing user needs and industry landscape.





