I see a lot of focus on improving efficiency and velocity with the usage of LLM’s. And it isn’t half bad. Alternatively, you can use the LLM’s and still move SLOWLY: HEAVY usage of AI tools, but purposefully moving slowly.

I retain slow velocity and remain “inefficient” on purpose. I remove myself from the implementation to focus heavily on architecture and design. In my eyes, the LLM’s allow us to trivialize things that, five years ago, take a long time. Now, we can instead increase the scope of our projects.

I like starting with a folder in VS Code. I create a primer markdown file that isn’t committed to the repo. Every developer on the project has their own primer. Mine is a Senior AI Engineer with 10+ years of experience. A junior developer has their own primer explaining their level, etc.

Then I create a project ideas markdown file, plus API documentation and JSON result examples.

I build up slowly.

I like using GitHub Copilot inline autocomplete to help me create these. You start typing, and it starts reading your mind. The primer and my own writing style in the markdown give the AI assistant enough to work on.

Then I have Opus do a pass on the project ideas to create a rough plan. This is like a rough draft.

I then utilize GPT-5.2 to make that plan readable for humans.

And then I establish a GitHub repo with a README.

I keep design, coding, and project time the same, and I just increase project scope instead of aiming for faster timelines and rapid velocity.

Nothing flashy. Slow, steady, and ready to win the race that matters.