[Jodavaho.io]

On the optimality of sawtooth careers

I was having lunch at JPL with their Chief Engineer, Rob Manning. Normally, you’d expect the Chief Eng. to reserve their time for more important people, but that’s not how JPL works.

I had posted in the New Researcher Support Group slack channel, asking for stories from folks who regret some part of their career. Gloomy, I know. He was curious, and so we we got lunch.

Being early career is fraught with uncertainty. What should I specialize in? How can I make an impact? etc.

He cut through it pretty quickly with his model for an optimal career: The sawtooth.

Rob said that he had a horrible character flaw, in that when he got bored of something he’d usually leave to do something else¹. This led to him perpetually entering new fields as a beginner, asking dumb questions and holding himself accountable to learning quickly. He said this breadth of perspective allowed him to become a far better engineer over time.²

I respect that. Tying it back, it helped me see that just doing something, and doing it fully, is better than not. If you hate it, at least you did it, and now you can go dedicate yourself to something else.

For that reason I was perfectly happy not applying for management positions when I left Amazon. Even though this is kind of a step down. Or, when I was a manager, I could get over it and do small little contributions to small little projects because I could learn something.

As a meta-cognitive skill, I’d like to become better at walking into new areas as a beginner, and less reluctant to do so. Less pearl clutching, more exploration.

In the last 3 years, my saw-teeth have been:

• Learning to optimize with SIMD, CUDA, and what you might call “esoteric problem reformulations”.
• Beginning projects in new languages. Zig, Rust, or (since I rarely use it anymore) C.
• Diving into backend development with both feet, especially on AWS and vercel.

These should be evident in projects.

And, going forward, I’d really love to do more:

1. High performance or low-latency computing
2. Image processing, especially for tracking
3. Work/research on parallelization for search, esp in game theory or discrete optimization

Now I just need to figure out how.