Last year I bought a remarkable. I've used it for writing and focusing my thoughts. Before I work on a system design or a feature, I'll often hand write the design. This ends up feeling like a "stream of thoughts" more than an organized spec, but it does just the same. When I do this, my results are far higher quality than if I shoot from the hip or ride the vibes. Thinking about this more, it makes sense: writing is thinking. We used to write code by hand (some of us still do, but not much anymore). And now we may prompt agents, but if you outsource your thinking, not only is this dangerous, but it is lazy.
When you write your system design or solution by hand, it does a few things;
- Typing is faster than writing (I type a lot faster than I write)
- There are less distractions (no computing device/web browser)
- It forces you to slow down. You can't hide from the problem.
- If you can focus your thoughts into "contiguous thought streams", it's easier to follow, and there is less ambiguity.
So in other words, you're basically naked. You'll quickly realize where your gaps are. If you're using coding agents, this matters a bit more. Context matters the most. If you direct it where a part of the architecture lives, it doesn't have to waste turns finding it, or tokens on understanding the system. And you also understand the system, so your feedback loop will be faster, because you need less turns.