If a project needs some thought (e.g. each step will take more than 2 human hours)

When you're working on a more complex project, treat the setup as part of the engineering work.

First, choose the model and effort deliberately with /model and /effort. A quick bug fix and a risky architecture change should not necessarily use the same settings.

Second, use /plan when the project needs research before edits. Planning mode lets Claude inspect the codebase and propose a path before it starts changing files.

Third, use /btw when you have a quick side question that should not become part of the main task history. It is useful for questions like "what does this acronym mean?" or "is this API deprecated?" while keeping the main context clean.

Fourth, split work across subagents or Agent View when one conversation would become too broad. Give each worker a narrow task and a clear handoff.

The key is matching your approach to the complexity of the problem. Simple tasks don't need all this machinery, but when you're tackling something truly complex, these tools give Claude the cognitive resources it needs to do the job well.