Twitter seems like magic: You send a tweet, and it’s sent to your many followers in something closer to real time. But of course, Twitter — like all network applications — makes use of millions of networks that are not under its control. Many of those networks are slow and unreliable. How can you create a network application that appears to be seamless and reliable, working with images as well as text, when it is built on such an unreliable, unmanageable infrastructure? In this talk, Jess Garms describes the ways in which Twitter has worked to increase not just the reliability of their systems, but also the appearance of reliability for their users.