I have a rule I use for fighting this: don't learn more than 1 new technology at a time. On any project there will be a bunch of technologies I'd like to use but am not proficient with. I limit it to just one at a time. Sometimes that leads to temporary scaffolding. For example, I want to do this server component in Elixir, but I'm also using a new client technology. So I resist the urge to do Elixir and do the server component in my old standby, Python, for now, and perhaps redo it later.

Generally you have to decide whether you're trying to create products, or create street cred. If you're working on creating street cred, you'll make better progress if you learn one new tech at a time.