Things That Don't Scale, The Software Edition – Dalton Caldwell and Michael Seibel

0:00 – 9:47

Gmail and Facebook: the products that started as glorified hacks

The video opens with a brutal metaphor: instead of inspecting every pipe in a house to predict which one will leak, turn the water on and see what actually breaks. That becomes the frame for everything that follows. Michael Seibel and Dalton Caldwell start with Paul Buchheit, the creator of Gmail, because he embodies the idea better than almost anyone. His rule was the "90 10 solution": how do you get 90 percent of the benefit for 10 percent of the work? Founders often hate hearing that because it sounds like cutting corners. But Buchheit had unusual credibility. He built Gmail as a side project and ended up creating something used by 1.5 billion people. What makes the Gmail story memorable is how personal and unglamorous it was. Buchheit was annoyed with the email product he had, so he hacked together an early version of Gmail by putting his own email into a Google Groups style interface. His real breakthrough was not a grand product plan. It was the moment he could read his own mail in the new UI and stop using the old client. From there, each feature was pulled into existence by immediate need. He wanted to reply, so he built sending. A coworker liked the speed but pointed out the obvious flaw: "it has your email in it and I wanted to have my email." So Gmail had to become multi user. Even reliability was discovered the hard way. One day Gmail went down, people panicked, and Buchheit walked to the server room with a screwdriver to deal with a corrupted hard drive. That was when he learned email is not a toy, people expect it to always work. Even the famous invite system was not some brilliant viral masterstroke. Google simply did not have enough server capacity, so limiting invitations was a necessity. Dalton then shifts from software lore to a firsthand memory from a Santa Clara data center, and the picture is wonderfully physical. Before cloud computing, founders literally drove to anonymous warehouse buildings full of racks, blinking lights, and overheated machines. There he noticed a new neighbor labeled "thefacebook.com." The servers looked cheap and messy, but each one carried the name of a university: Stanford, Harvard, and more. That revealed the trick. Facebook had effectively cloned itself school by school. Each campus got its own PHP instance, its own MySQL database, and its own cache. Instead of solving the impossible problem of one giant database for millions of users, they created many small islands. The product looked unified from the outside, but underneath it was a copy paste empire. That is the unexpected angle of the whole section: many legendary companies did not begin with elegant architecture, they began with whatever ugly structure let users in the door.