Life Update: Just Moved to SF Bay Area
TL;DR: I moved to the Bay Area from Taiwan with the family to work on Recce.
It was on our drive back from the hotspring in Taiwan when my wife Ipa said, “If this is your SVK moment again, there’s no reason not to move to SF.”
Here we are, with lots of help from friends to navigate the move.
SVK
SVK is the distributed version control system I developed about 20 years ago. It was one of the pioneering systems exploring today’s software collaboration. It was shipped with macOS for a few years, used by gaming studios and Apple internally, including the WebKit team until recently.
Two years later, Linus created git, then GitHub was founded in 2008, and the rest is history.
The Decision
When I returned to Taiwan from SF in late 2023, the first time since the pandemic, I talked to my wife about moving there. We both lived in the UK before returning to Taiwan. I thought my expat life was sorted, especially after having kids. But the SF trip and reconnecting with friends who moved there made me feel “omg this is my crowd.” And for a devtool startup like Recce to work, it has to be there!
SF is a unique place where unreasonable positivity fuels wild imaginations into reality. It’s the best place to make mistakes and learn quickly.
Recce
What am I working on now? We’re solving the bottleneck for the future of software development. A few things I observed:
Code-gen is not a bottleneck anymore. The new bottleneck will be code review and deployment decision.
Correctness is often subjective with data-centric or AI-centric systems. Unlike traditional software, we assume correctness is verified by tests when reviewing pull requests.
The software development process becomes more experimental, and we need to use the pull requests concept differently.
In traditional software, creating a PR and passing tests gives high confidence with high test coverage. Code review focuses on code quality and edge cases.
In data systems, we haven’t established correctness even if every boilerplate check in your PR passed. Efforts in data observability and unit testing are either too late or require lots of activation energy.
We’re building the first version of Recce, focusing on the code review workflow for data-pipeline-as-code systems like dbt or sqlmesh. It allows authors to curate proof of correctness and stakeholders to review in high-level checklists.
It is still early but we’ve seen great interests and adoption of the Recce workflow in a few teams. I’ll share more about this journey and learnings.
So here’s the new chapter of life and I’ll share more about this journey!
I want to thank many people for helping with the move: Jos Boumans, Jesse Vincent, Tyson Mao, David Riordan, Colin Megil, Deger Turan, Artur Bergman, David Glasser to name a few.