My optimization principles

We brush our teeth in the morning. When we wake up, these tasks are automatic. These automated movements, refined through decades of practice using the feedback loop of pain, have led us to use barely any mental effort to accomplish this morning ritual.

This ritual, while recommended, frees up time and also leaves us with the mental capacity to do useful work for the rest of the day. We can draw an important lesson: we want to protect the limited amount of consciousness each day and allocate this toward useful work only.

It begins with the very first principle of optimization: turning conscious effort into the subconscious with a streamlined, no-effort-requiring process, just like brushing our teeth. Only then can we truly utilize our consciousness for creative and productive work.

Principle 1. Use minimum input to attain the same output

The ideal goal is to get the work done without unnecessary physical movements. For example, if you have 10 issues and pull requests to comment on, you might want to type “lgtm” to save time and physical effort, but this level of brevity should be used carefully. Hence, we recognize this gap.

I begin by recording relevant shortcuts in the IDE, browser, GitHub, terminal, and operating system. Here are some of the shortcuts I use daily:

  • Use a text expander to avoid typing the same text repeatedly. I use espanso to create text shortcuts. See text-expander-with-espanso for more details.

  • In a browser, switch between tabs using cmd-1, cmd-2, etc., and cmd-[ and cmd-]. Create a new tab with cmd-t and close the current tab with cmd-w. Use cmd-shift-t to reopen the last closed tab.

  • On GitHub, visit the notifications page, issues, and PRs using G-N, G-I, and G-P respectively. Create a new issue with C on the issues page. Create a new PR with C. See all the shortcuts by pressing shift-/ on github.com and refer to the Keyboard shortcuts page.

  • On macOS, use the Magnet application to manage and arrange windows without the need to use the mouse, splitting apps onto the screen and maximizing them using cmd-< for the left half and cmd-> for the right half.

  • In the IDE (e.g., Visual Studio Code), use cmd-t or cmd-p to search for files, cmd-shift-p to search for commands, and cmd-, cmd-1, etc., to switch between tabs. Determine the most effective ways to navigate lines, cursors, files, and to delete or modify content.

  • When needed, set custom keyboard shortcuts to reduce strain on your fingers and wrists.

I also create my own shortcuts with custom alias commands. To write this content, I woke up and simply typed bweb; it (a) navigates to my bobleesj.github.io project path, (b) activates the relevant conda environment, (c) displays the latest commit messages, and (d) opens Visual Studio Code. To learn how to set up keyboard shortcuts in your terminal, see Keyboard shortcut.

Principle 2. Attain maximum output with the same input

It is common for open-source scientific packages to release a package using the git tag <version> command to tag the release version on GitHub. In scikit-package, I led the development of a reusable workflow so that uploading the tag also checks admin privileges, releases on GitHub, uploads code to PyPI, updates the CHANGELOG, and deploys the documentation.

For my daily work, I use the pomodoro technique to maximize my output, with 50 minutes of focus and 10 minutes of rest in between. I have found that this is the most optimal way to maximize my output instead of working many hours straight without breaks. You can read more about my pomodoro technique described in My approach to daily work.

Principle 3. Minimum process with the same input and ouput

There is a function between the input and the output. A simple function means less maintenance and development effort are required. For example, we have Python version 3.13 hard-coded in our scikit-package GitHub Actions workflows. The simpler it is to update these Python versions across multiple files. This is precisely why we have adopted the reusable workflow in scikit-package, where we can update these Python parameters in one place rather than modifying them across dozens of projects.

time is an important part of the process. Instead of conda install, mamba install is preferred because the time is reduced despite the same cognitive and physical input and outcome. I find the Brave browser faster than other browsers and block ads by default, during the process which my time and consciousness could be used for not useful work.

Another example is the use of vectorization for computing distances between two points within a unit cell and its supercell as implemented in cifkit. Using the for-loop method is often the first approach that comes to mind, but it is not the most efficient. Instead, I had a collaborator who made a significant contribution that boosted the algorithm’s performance more than 10x. This is an example of how we can use the minimum process to achieve the same output with the same input.

Takeaway

Principle #3 is often intertwined with Principles #1 and #2. Using the minimum process may also mean using the minimum resources, such as electricity and time, to achieve the same output (Principle #1). However, I wanted to separate this principle from the first two because it is the perspective-input, output, or process—that we are trying to optimize. Each workflow may have different priorities. It is possible to apply all but it’s good to divide and conquer optimizationz

Exceptions and questions

There is one exception when the principles should be violated: when |effort required for optimization| is greater than |total effort saved| over the entire lifecycle of the workflow.

Here are some questions I ask when I am about to embark on a new task or workflow:

  • Which principle can I apply to this workflow? (Choose Principle #1, #2, or #3)

  • How can I reduce the effort required for the same task? (Principle #1)

  • How can I maximize impact with the same effort? (Principle #2)

  • How can I simplify the automation process? (Principle #3)

*First draft on Jun 24, 2025 (Incheon, South Korea)*ork