Skating to Where AI Is Going, One Pull Request at a Time

Last week I submitted my first ever pull request. Yes, you read that right, a UX Researcher submitted code for review to become a part of a software codebase. I know this may not be earth-shattering news to some, but for me it’s a big deal. In my previous post I wrote about my quest to become a builder, and I just started making significant moves to make that a reality. 

I have no intention of leaving research behind. Rather, my goal is to be a researcher who can ship the fix by closing the loop between identifying a problem and delivering the outcome.

The market is moving

AI is a flaming puck moving with high velocity in today’s world.

In 2025 the Professional Women's Hockey League expanded into Seattle with the Torrent. My family and I serendipitously attended their inaugural game of the season and I was instantly hooked. Before I knew it I was attending more games, bought a Hilary Knight jersey, and was emotionally distraught with the draft outcome during the off season because we lost so much talent. Congrats Detroit, you played your cards well.

I bring up hockey in a blog post about AI, healthcare, and UX because I think it’s going to be prudent to anticipate where the domains are moving, rather than just focusing on augmenting existing workstreams and processes. AI isn’t going anywhere, and once the push to gain first mover advantage in greenfield markets subsides, the focus on quality UX that creates material business value will become the primary battleground again. 

Recent headlines have highlighted how AI costs are ballooning much higher than expected. Uber spent their entire AI budget in the first four months of 2026 while questioning whether the productivity gains justified the cost. AI spend also continues to climb despite decreasing inference costs. In addition to those trends, frontier labs have begun laying the groundwork to IPO, which means they will need to become profitable to deliver shareholder value. 

This pattern isn't new. Global energy use for space cooling more than tripled between 1990 and 2016, even as the underlying technology kept getting more efficient. When the unit cost of something useful falls, people don't buy the same amount for less, they buy more.

Furthermore, we’re starting to see signs that the era of tokenmaxxing is coming to an end. Companies are taking down AI use leaderboards and asking employees to use AI more meaningfully with the associated cost in mind. Anecdotally, I’ve been saying to friends and coworkers for a few months now that I used to take Lyfts to/from the airport, now I park my car at the airport if the trip is less than two weeks because it’s cheaper. Looking ahead, my hypothesis is that when every token has a price attached, the experiences that win will be the ones that get users to the right outcome the first time. Which means “right” stops being a matter of taste and becomes something you have to define and measure. That's a research problem before it's a design one.

Getting back to my healthcare roots

A couple of months back I was catching up with my buddy Piotr Mankowski, where he was showing me some of the work he was doing in support of the open-source electronic health record system, known as OpenMRS. OpenMRS traces back to a 1990s medical training partnership between Indiana University School of Medicine and Moi University in Eldoret, Kenya, which shifted toward HIV/AIDS care as the epidemic swept western Kenya. By 2004 the patient load had outgrown the program's Microsoft Access database, prompting the Regenstrief Institute and Partners In Health to collaborate on a replacement. They built it in the open, and it became the world's leading open-source medical record system. Currently, OpenMRS is supporting human health with 5,000+ community members, across 8,100+ facilities, in 80+ countries, and 22 million individual patient records as of mid-2026. 

Piotr has been working to integrate LLMs into OpenMRS. My interest was immediately sparked by the project because in 2019 I successfully defended my dissertation to earn my PhD in Health Informatics from the University of Washington School of Medicine. My research involved designing, building, and evaluating an AI agent called Hernia Coach to answer common questions from surgery patients undergoing treatment. You can learn more about the project on my website. I have felt a calling to begin contributing to healthcare again, and I saw there was a substantive opportunity for me to be a part of the efforts.

The OpenMRS community has gained one more contributor. I submitted my first pull request to improve locally running the system on a personal laptop by streamlining the start/stop processes with scripts that can be invoked like an application in the dock rather than requiring the use of the terminal. It's a milestone toward becoming a builder.

What comes next

I’m going to explore a full-spectrum approach to my involvement with the OpenMRS project. First, I’m going to assess the current quality of the user experience of the LLMs. Second, I’m going to build iterative improvements for the OpenMRS platform based on my assessment’s findings. Third, I’m going to learn how to run automated evaluations that assess LLM quality at scale. 

So I’m curious, where do you think the puck is moving in your world? How do you think quality is going to be critical for success as the push for first mover advantage is less of a priority?