Developing an algorithm is not the hard part. Adapting it to customer data and wishes, is.
If/when Twitter launches an algorithmic timeline, it is probably going to suck at first. Only by measuring huge volumes of user interaction with the feature, will they get enough data to make it work well.
I mean, look at this thread--people are praising Facebook's ease of use, but back when Facebook starting filtering their timelines, a) power users were super pissed, and b) a lot of random stuff showed up. For a long time there was a semi-hidden "show me all posts" option for the timeline, which power users turned on. That doesn't exist anymore, and most people don't miss it, because the filtering is good enough now.
There's not really one algorithm. There's a learning system that creates, on the fly, a billion custom algorithms--one per user.
Learning systems need two things to work well. They need a huge corpus of data, and they need clear guidance on the desired outcome. Twitter has a lot of tweets, but they won't really have clear guidance until people start actually using the feature.
I'm sure they're trying to collect some now, with their "while you were away" feature, and tracking interactions like you unfollowing Bob after seeing that tweet from Ann.