Start playing in a domain as closely related to the work as possible (i.e. explore a tangent you're curious about), then, if/when you gained some momentum try to identify a good stopping point try to smoothly pivot into what you should really be doing instead of stopping.
It doesn't always work, but every now and then it does.
Sure, the end stage is to work, but the first stage can be anything. Call a friend, ask a coworker to help, or peer programming, or look at the description of the task, keep a journal [like lab notes], write pseudocode, open an empty document and break down the task into smaller ones. add emojis if you want to to keep track of progress
It’s an important skill to be able to reframe things for your brain, and not just in this context.