Minimal Viable

“Huh.” It wasn’t the response I had expected. Emery Brown’s plenary at ISIT 2007 was one of the highlights of the conference, and seeing someone apply a state-space model to predict a mouse’s movements from neural spikes was inspiring. I thought my excitement in relaying it would be infectious. That this Huh was coming from someone in Brown’s lab made me wonder what I had missed.

“Did you notice that naive linear regression performed almost as well as the state-space model?” And there it was. I had been so caught up in what Brown’s team had built that I didn’t really care how they built it. They could have demonstrated the result much earlier without relying on the state-space model, and I would have been just as impressed.

This idea ended up working to my benefit multiple times during my time at Google. In a company with a short attention span, demonstrating things quickly has its advantages, and that often meant leaving more sophisticated approaches on the table to get to something that worked.

However, those more sophisticated approaches were more exciting to work on, and I would often find myself or members of my team drawn to them. New technologies, techniques, and approaches would call to us, and we would be taken in. It required a certain discipline to prioritize what was important to get to a minimal viable solution.

And so it’s been with my life outside of Google. The pace of technology has generated a trail of methods with a strong allure, and my first couple months have included days in which I was drawn in by ideas that have introduced more challenges: difficulty debugging, complex undocumented dependencies, and expensive resources to run.

And then there are days like yesterday, in which, shortly before bed, I wrote down a technique that I would be able to validate in a day, relying primarily on textbook approaches with well documented libraries, debugging tools, and best practices. Also, it could run on my local CPU. With a few hours of effort today, I had an end to end system that mostly worked, or at least provided a baseline.

Minimal. Viable.