In some cases, you'll have a specific client that wants something specific and you'll just have to make your best guess. But, at a lot of companies, there's no specific client who needs feature x by day y. Estimating in those companies is almost always a waste of time, a nonsense exercise designed to make managers feel like they're in control of the situation.

I come to work at the same time every day for the same amount of time every day and I work on our product that whole time. What's going to get done is going to get done. If a situation requires more, then we'll work late. (And it's almost always obvious when something is important enough to require this.) The process of guessing how long each task is going to take is a complete waste of our time. Of course, we do it. And management thinks it's super important, because they spend their days making spreadsheets and Jira filters based on those guesses. But, the reality is that it simply doesn't matter one way or the other.

We make a bunch of guesses, then we do the work we were going to do anyway and it takes as long as it was always going to take, and then at the end we have another pointless meeting, called the "retrospective," where we talk about how good or bad our guesses were.

It's like a Tennessee Valley Authority make-work program for under-challenged product managers.

Think of a video chat feature for a code pair website. Maybe you can deliver it faster if you build a central server based video chat model using a standard VP8 than if you go WebRTC and a heavily customized codec.