A good toolchain should shorten the distance between thinking and implementation. The number of tools is not the goal; reliably completing a real task is.
Think: write the problem down first
Before opening a chat, write down the goal, known constraints, and what success looks like. Let AI propose multiple paths, then choose deliberately. Model output is a candidate solution; important facts still need to be checked against primary material.
Build: keep context inside the project
Keep code, design notes, and run commands in the same repository. Make each change concrete: one problem, one implementation, one verification step. That lets any coding assistant continue from the real project state instead of reconstructing context from memory.
Organize: leave records you can reuse
Markdown preserves reasoning, Git records change, and a static site turns those artifacts into searchable knowledge. Add new tools only when the existing workflow has a real bottleneck.
The simplest way to evaluate a tool is to use it for the same small task: record the time, verify the result, and ask whether you would choose it again tomorrow.