Google's Director of Technology Talks

Craig Silverstein from Google, employee number one and director of technology, is talking here at Emerging Technology about how Google accomplishes innovates, rather than how they technically carry out their tasks.Craig says that the goal is to do things that matter. He used to believe that "for a company to be successful, it really had to be evil." But Google has proven that you can do things that make a difference to people and succeed. "It's one of the reasons we're still around after the bubble." Google resisted pressure to run popup ads in the early days, as well as banner ads, because they wanted to create lightweight unobtrusive text ads. "It's successfully bringing in money for us now." At Google Labs, the company puts out ideas that they aren't quite nailed down on yet. They don't know the precise feature set or how to accomplish a certain goal. The feedback allows them to "do something that's really useful to people" when and if they turn it into a product. The cost of switching search engines for the user is zero. "We have to have it that users would rather go to Google than other search engines just to survive."' Craig emphasizes hiring: "it's the key to what makes our process successful." His slide reads, "brilliant people have good ideas." Google trusts that its people will do the right process because they've hired the right people. "They realize the value of trying to do the right thing." A creative environment creates creativity. Design is a first and primary component of their development process. "Our site looks simple, and looks artless almost, but it takes a lot of work to get there." He shows the first beta search interface for Google, which he says an article described as "anorexic." Hard to get things wrong with a single form, and fast to download. They counted the bytes on the home page, and they still do. Google keeps "the top 100" list: the top projects that they'd like to do. It's not 100, might be 130. They're not working on all of them, but by keeping this list, and ordering them vaguely by priority, they can keep their mission statement in mind -- they don't wind up working on something unrelated. He shows a page of the list. The first item is, naturally, Build a Search Engine. Number two: Crawl the Web. Number three; Google News. "Just last month we got the idea of building a search engine, so we'll be working on that." (laughter) There's an iterative process of figuring out what works. Shows a chart of wireless (cell phone) traffic. It meandered for a while, and then jumped when they signed a partner last Christmas. It was way down on the top 100 list, and now it's maybe number 12. They meet weekly and review priority of the "living document." "It's very much essential for this to be a success." Their approach is to use small teams: "small teams are fast and agile," the slide says. The whole group works on a project, including design, testing, launch, and continuing engineering. The Web crawler team is Kingson and Jill. They crawl 3 1/2 billion pages per month, and "these two people are the ones responsible for making it happen." Other teams help, but they own it. Hierarchical communication is antithetical to their small group approach, so they have to constantly ensure that communications continues across the entire organization. When a project is ready to start, the entire company can be involved, but a few people are responsible for dealing with comments. "Make sure the groups are sharing experiences, sharing technology, sharing ideas." They have an interteam group, which is really 10 groups of three, not one group of 30 people. One group helps make sure that new people can understand how to get into the code base. These positions are "volunteer": people take time out of other projects. They have a weekly tech talk in which people talk about what's in progress or finished, and then they put the talks up on the Web site as videos. It helps new employees understand what happened with the project. Google creates tools to help them foster communications. They have an internal search engine -- "you can guess which one" -- over their internal Web sites. They have a status page that's created from emails sent to it and that maintains state for a week. When Google acquired Pyra Labs (Blogger's creators), none of the press speculated that they might use blogs internally, but that's the first thing the group suggested when they came on board. In the old days, they put stuff up and waited for comments. But now they do user studies. Their first stab post the initial home page was, "I'm waiting for the rest of it." "Is this some guy's home page?" "How many people work there?" Google Labs allows them to test ideas that aren't ready for prime time. "My particular favor is Google sets." You enter the first part of a set, like the names of some of the Seven Dwarfs, and it helps complete them. Google.com/jobs: "I encourage anyone who is thinking about switching jobs or starting a new one to apply to Google." They hired a compiler expert, which they didn't exactly need, but the first thing he did was massively improve the speed of part of the back-end. They have one hiring committee, just a few people, and this committee hires everyone. "It manages to keep us consistent" and it separates hiring from head count. Someone might say we need 10 engineers to make the ad system up to date, but that person isn't making the hiring decision: the committee doesn't have those pressures. "It's much much more important to use that we hire people who fit this criteria we have than to miss people who are great." "Bad employees are the time sink that keep companies from being as successful as they could be." It allows them to run all the projects they do at the same time without worrying about these hiring issues. Only a few people at Google haven't worked out. A Google News home page can't be lightweight: sort of the opposite of news. They wanted something that was "googly." He walked through an iteration and how they decided on several elements, including top stories. During the bubble, people said, "The Web changes everything." That true? Not exactly. But before the Web how would they have created what they do? Google wouldn't have existed, of course, but before the Web, simple search engines with unsophisticated search engines worked fine. Wouldn't be able to have small groups because you couldn't talk broadly enough efficiently enough. Sharing information on Web pages that archive information "that allows us to have the kind of communication we need to have these small teams." The company runs the way you think you might run a company if you weren't actually running one. Especially unlimited food. (I visited Google yesterday, and their biggest problem in scaling has nothing to do with their search engine, but rather with their dining room. The Grateful Dead's former road chef runs the free restaurant, serving lunch and dinner, and they've got so many employees at Google now, that even with an overflow outdoor tent, they can't quickly handle the full-on lunch load. And, man, is that food good.) Q: With new office in New York, how do you maintain culture? A: Groups there working through same processes. Q: [inscrutable question on groups] and then why not more than 3.5 billion pages. A: I didn't understand your first question, so I'm just going to answer the second. Some finite resources, but will be solved over time. Q: Applied Semantics acquisition? A: Can't comment directly, but name implies what it does. Q: How do you rein people in? A: Happy to have unrelated ideas, although they might not get selected to discuss. Gatekeepers are technologists who stand at different parts of the process to see what's considered. Q: RIAA sued students who built search engine for college intranet. Google is better at searching MP3s...if you type "MP3s and..." Craig: Stop it! Right there! [laughter] A: Why don't we have a music search? Exactly this reason: intellectual property. Not much legal music and video on the Web, and "we don't want to make it easy to find this." Can only imagine illegitimate uses for it, then won't do it. But if it has a good goal, they don't try to secondguess. Only nix pages that the content owner asks for or that are stolen.