I can feel how quickly the tool stack becomes a way to avoid the harder question.
Before the work has a shape, it is very easy to start designing the stack: which app, which agent, which database, which automation. But the first useful move is usually tuning the conversation: the project structure, role context, instructions, and the way the assistant should challenge, remember, and help think.
The stack still matters. But without a good interpretive layer, every new tool just creates another place for context to scatter.