The policy is that you don't ask, you tell. If your manager needs you to focus on your primary project, she can ask you to bank your 20% time for up to a quarter, but then you get to use that saved time.
I'm a manager at Google and I think 20% time is great. Engineers explore something that's exciting to them, and the projects usually pay dividends, either in great launches or else people get a chance to learn new skills or hone their current skills.
It entirely depends on your manager and team. Some managers might want you to ask permission, and/or if you take the "forgiveness" approach your relationship with that manager might suffer. But that's because people are people.
I can't imagine my former TLM ever taking 20% time, but I can thoroughly imagine him encouraging every single one of his direct reports to do so, unless we were on a launch sprint.
Frankly, I don't think that inconsistant application of rules, and relying on a good Manager is a good thing overall.
I don't want to equate this with working at Google, because I don't have enough information, but here's my anecdote about good managers:
Twice before I've worked with spectacular managers. They treated their employees well, and acted as great shields against the political infighting inside the company. But both times, the managers were forced out, and the employees that depended on them to get interesting work done ended up quitting as well.
If this is indeed, as the TFA posits, a memo from the top being filtered out by a few good managers, in a short amount of time those managers won't matter; they will be forced, or burnt, out. Neither case is good for the company.
> Frankly, I don't think that inconsistant application of rules, and relying on a good Manager is a good thing overall.
If you need humans instead of robots doing a job, that usually (increasingly, as automation advances) means it requires judgement such that pre-written inflexible rules will be inadequate to handle it sufficiently. Which means you need to rely on the judgement of people applying flexible rules for the best results.