Learn, Build, Repeat

Almost exactly a year ago, I stood at the front of my living room at our quarterly partner planning session emphatically declaring, “no, really, people do this manually and it’s a great use case for genAI… seriously we need to build this”. The “this” I was referring to was, of course, the process of decomposing mortgage regulatory change obligations into developable units of work (requirements, user stories, acceptance criteria, and test cases was how I thought of it at the time).

I was dead certain about the problem and a little fuzzy about the solution. Ok, ok, fine… I had only the haziest idea WTF the solution was, but I was sure we could figure it out. It was a good thing we had absolutely zero AI or machine learning experience because we needed to hurry up and finish immediately. I was convinced we were very, very late. I was certain literally “everyone else” had figured it all out. ChatGPT came out in November 2022 and here we were in December 2023, having never even logged into the damn thing. Time is never our friend.

Lucky for me, it wasn’t all figured out and I wasn’t (that) late. In fact, it’s still not figured out and this journey of genAI is the wildest technology ride I have ever been on. Time is still not our friend, and I am hanging on for dear life as we bring our product to the general market after our rollout to our first set of internal users. I just finished watching Yellowstone and I feel like the genAI journey is like Jimmy trying his hand at rodeo… and just barely surviving… and then trying again… until he finds his way to the four sixes ranch. I haven’t gotten to the four sixes ranch yet.

Many of the concepts and visuals from those early days do still resonate. And it is fair to rethink previously unimaginable ideas because we can now do things that would have been impossible before. Sure, I’ll just take the entire Fannie Mae guide and turn it into all its requirements, user stories, acceptance criteria, and test cases. That would have been insane before genAI. Totally doable now, expensive, but doable. So there really are no crazy ideas anymore.

We did dozens and dozens of interviews, still dozens more conversations, and got a really wide perspective on the problem as well as how others in and outside of mortgage were using genAI. We matured out thinking, deepened our understanding of the problem, learned the technology, and iterated our product, testing with potential users all the time.

In addition to doubling down on our problem, we doubled down on the technology. It was comforting to see so many others facing the same struggles. My team has grown in leaps and bounds, we are expert genAI builders and prompt engineers. I have also personally been deep in learning and applying mode. I’ve basically spent the past year learning about genAI, figuring out what it means to us in mortgage, and then applying the learnings within the team through building. I am so grateful to have had this opportunity, Learn, build, learn, build, that was 2024.