In the Plex

By Steven Levy

9 min read

Why does it matter? Because the company that set out to prove you could get powerful without getting evil became the most instructive case study for why that's harder than it sounds.

Halloween 1999: a Newsweek journalist sits across from a Viking and a cow in a Mountain View conference room. The Viking is Larry Page. The cow — rubber udders, full commitment — is Sergey Brin. They are explaining how academic citations are secretly a map of the entire internet, and why that insight means they can rank every page in existence by how much it deserves to be found. Ten years later, twenty-seven of their natural allies — libraries, authors, civil liberties organizations — stand before a federal judge calling Google a monopoly, a threat to free expression, an instrument of concealment.

The Viking and the cow were completely sincere. So were the critics. This is the story of how one company can earn both verdicts — and why an engineering culture that genuinely believed in doing no evil was the last to notice when it stopped.

Your Competitors Knew About PageRank — They Just Couldn't Afford to Use It

George Bell, CEO of Excite, watched from across his desk as two Stanford grad students typed a single word into competing search engines. The word was "Internet." His company's results surfaced a tangle of Chinese web pages where the English word appeared like a flag in foreign soil. The Stanford system (BackRub, built by Larry Page and Sergey Brin) returned pages explaining how browsers worked. It was exactly what someone searching "Internet" would want.

Bell was visibly upset.

Not because the demo failed. Because it worked. If Excite adopted this technology, users who found the right answer immediately would close the tab and leave. Bell's entire business ran on "stickiness": time users spent on the site, generating ad impressions. A search engine that solved queries on the first try destroyed his revenue model. He told the students he wanted Excite's search to be 80 percent as good as the competition's. Not better. Eighty percent.

That scene reframes Google's origin story. We tend to tell it as a triumph of two brilliant outsiders who saw something the establishment missed. But Bell saw it. He understood exactly what he was looking at, and chose not to use it.

The insight Page and Brin had developed — that web links worked like academic citations, with the most-linked pages gaining authority the way the most-cited papers did — wasn't even unique to them. A computer scientist named Jon Kleinberg independently reached the same conclusion at IBM, only to watch his managers shrug: a $70 billion company couldn't see the business case. A Chinese engineer named Robin Li, who would later found Baidu with this exact idea, filed a patent in 1996, handed it to Dow Jones, and was ignored. The idea was everywhere.

What Page and Brin had that Bell didn't was the luxury of wanting search to actually work. They had no stickiness to protect, no investors demanding they keep users trapped on a portal. Whatever business model they eventually found would have to be built around search being as useful as possible. Bell's had to be built around the opposite. He didn't make a mistake that day. He made a perfectly rational business decision that happened to be catastrophically wrong.

The Engineer Who Hated Ads Built the Most Successful Ad Machine in History — by Treating Relevance as a Moral Standard

The quality score encoded a single belief: relevance is a financial obligation.

Eric Veach, the Canadian mathematician who built AdWords, said he hated advertising. He'd won math olympiad prizes, helped render Pixar films, and arrived at Google in 2000 to find himself assigned to a seven-person ad team nobody else wanted. His contempt wasn't aesthetic. It was structural. Under impressions-based pricing, advertisers paid to show ads whether anyone found them useful or not. Nothing in the system punished irrelevance.

The quality score changed that. When advertisers bid for keywords, their bids determined only half their ranking. The other half was Google's prediction of whether users would actually click the ad. A high bid could be outranked by a lower one backed by stronger relevance. A sufficiently irrelevant ad could be priced out entirely.

The hand-lotion example shows how it worked. Three advertisers bid on that keyword. Alice sells artisanal lotion and bids 10 cents per click. A drugstore owner bids 15 cents. A travel advertiser bids 50 cents — he wants to reach the kind of person who buys hand lotion, not sell lotion himself. Under impressions-based pricing, the travel site wins: highest bid, top placement. Under AdWords, it finishes last. Google's models estimated that users searching for hand lotion wouldn't click a travel ad, so the 50-cent bid was discounted into irrelevance. Alice's 10-cent bid, backed by a high predicted click-through rate, secured the top slot for less money than either competitor paid.

Sheryl Sandberg ran Google's ad operations and didn't recognize what she was looking at until she called Treasury Secretary Larry Summers. The mechanism turned out to be the Vickrey second-price auction — the design that won William Vickrey a Nobel Prize. Veach had reinvented it independently, working from his conviction that winner's remorse was structurally corrupting auctions. The consequence was simple: advertisers now had a financial reason to fix their keywords, their copy, their landing pages. Relevance wasn't optional. Ignoring it raised your costs.

That's the mechanism Veach built from his hatred of bad ads. Most companies posted "don't be evil" somewhere and called it culture. Veach encoded it into a pricing formula, one that turned out to fund everything Google built for the next decade.

'Don't Be Evil' Began as a Joke at Intel's Expense — and That's Precisely Why It Worked

Paul Buchheit sat in Charlie's café on July 19, 2001, watching Stacy Sullivan write corporate values on an easel pad. "Play hard but keep the puck down." "Google will honor all its commitments." At Intel, his previous employer, every employee carried a laminated card listing company values on their badge, so when something questionable came up, you could look down and say "this violates value number five." The whole exercise was performance: words designed to make management feel responsible without changing how anyone actually behaved.

So when his turn came, he said it partly as a joke, partly as a dare: "Don't be evil."

Sullivan wanted something positive — "Do the right thing." Marissa Mayer, who ran search quality, agreed. But Buchheit and engineer Amit Patel held firm. The logic was simple: if you need to print "do the right thing" on a badge card, the behavior has already failed. "Don't be evil" worked because it named the temptation instead of pretending it didn't exist.

Patel then turned the phrase into infrastructure. Working alone, he wrote it in a calligraphic hand (Tolkien-esque, someone said) across every whiteboard in the Googleplex, until communications director David Krane called it "atmospheric, pervasive." When a meeting produced an idea with a whiff of anticompetitiveness, someone would note it sounded evil. End of idea. The phrase worked as a shared conscience. No enforcement mechanism needed — it was already in the room.

The test came three years later, at the IPO. Larry Page embedded the phrase in his letter to shareholders inside Google's S-1. The SEC asked Google to remove it. Google refused.

That refusal is the important detail. The easy move was to comply — strip the idealism from the legal document, keep the motto internal, protect the offering. Instead Page left it in. The IPO had already antagonized Wall Street at every turn: Dutch auction, dual-class shares, a Playboy interview mid-quiet-period that nearly derailed everything. Brin and Page each ended the day worth $3.8 billion.

The phrase had traveled from Buchheit's contempt for laminated badge cards to federal securities law. It held.

Google Used Its Own Founding Logic to Justify Becoming a Surveillance Tool

For years, Michael Brin produced lies for a living. A mathematician at the Soviet government's planning agency Gosplan, he spent a decade generating reports proving that Russian living standards surpassed America's. He knew they didn't. His Jewish background had already blocked him from an astrophysics career and redirected him to economics. He applied to emigrate in 1979, a decision that brought police visits to his apartment. His teenage son Sergey, taken back to Moscow years later, turned to his father after two days and said: thank you for getting us out.

That backstory explains why Google's China experiment ended when it did. Not through a business calculation, but through a personal reckoning.

The entry decision had been entirely analytical. Eric Schmidt described it as running an "evil scale": weighing the cost of censorship against the benefit of giving more than a billion people access to Google. If you accepted that framework, the math seemed to favor entry. Andrew McLaughlin had spent nearly a year on the question and came out against it. Bargaining in good faith with people who weren't, he argued, was its own kind of corruption. He still lost. Page believed technology would open China over time. Even Brin, despite his history, went along. They would censor selectively, tell users when results were blocked, and trust that the information they did provide would slowly erode the government's grip. Google's own founding logic, that information access is a moral good, justified the compromise.

What that logic couldn't account for was where it led.

Just before Christmas 2009, Google's security team traced a sophisticated breach to Chinese state-linked hackers. They had exploited a Microsoft instant-messaging vulnerability to reach Gaia, Google's central authentication system, stealing code so sensitive the company never disclosed its nature. That was damaging. What came next was worse: forensics showed the attackers had also accessed Gmail accounts belonging to Chinese dissidents and human rights activists. Not randomly — they had targeted specific people. The contacts, plans, and private communications of individuals the government wanted to identify had moved through Google's systems into the hands of those searching for them.

Brin's response surprised his colleagues. He was less disturbed by the stolen code than by this. Google had entered China telling itself it was bringing information to people who needed it. It had ended up carrying information about those people to a government that could make them disappear. The utilitarian calculation had inverted — not through any change in the math, but because Google had misidentified what it was actually building.

He presented the case for stopping censorship immediately, with what insiders described as utmost passion. Schmidt disagreed. Page sided with Brin. The decision was made January 10, 2010.

The evil scale moved in the end not through logic but through the fact that someone at the table knew from childhood exactly what it looked like when a government used a bureaucracy to find the people it wanted to silence.

Google's Test Population Was Twenty Thousand People Who Shared Almost Nothing with Its Users

How does a company obsessed with A/B testing, user data, and iterative refinement miss something that becomes obvious within 48 hours of launch?

The Buzz launch in February 2010 was methodical. The product had been tested for months inside the company. The night before release, the team ran a full rehearsal — product demos, feature overviews, details from each team lead. Then Google's PR staff pretended to be journalists firing their hardest questions. Nobody asked about privacy.

They couldn't — not because they were negligent, but because the question required imagining a user whose life looked nothing like their own. Buzz worked by auto-generating a social network from your Gmail contacts, the feature the team was proudest of. For twenty thousand Googlers, most of whose email contacts were other Googlers, this was genuinely useful: your professional world, instantly organized. When they tested it internally, they loved it.

Within two days of public launch, a domestic violence survivor discovered that Buzz had made her contact network visible to anyone who looked at her profile. Her abusive former partner could now see whom she corresponded with most, which revealed where she was living. Privacy regulators from ten countries co-signed a letter of complaint. Google's own former policy chief, writing on Buzz itself, said the default exposure violated basic expectations.

Google's chief privacy lawyer, Nicole Wong, named the problem afterward: "The on-boarding process is not like doing it in the wild, and the social network of 20,000 Googlers is not like being on the Internet. That process failed us."

Wong's diagnosis was structural. Google's internal culture was so professionally homogeneous (the same employer, the same educational background, the same Bay Area networks) that its test users shared almost nothing with the people who would actually use the product. The company had built the most sophisticated measurement apparatus in the industry and was pointing it at the wrong people. The harm Buzz caused was structurally guaranteed: a product shaped by twenty thousand people who had no abusive ex-partners, no reasons to keep their social networks hidden, no life circumstances where automatic contact visibility could end badly.

The Question Was Never Whether Google Is Evil — It's Whether Anyone Gets to Check

Imagine a water utility that decides, without oversight, which neighborhoods get clean water. Its engineers genuinely want people to have water. Its models are sophisticated; its water mostly good. But when someone challenges its decisions, the utility points to its own models as proof it got it right. You can't audit the models. Trust the utility, or go thirsty.

That's the structural position Google arrived at. By design. Larry Page's argument during the Book Search lawsuit crystallizes the pattern: "Do you really want the whole world not to have access to human knowledge as contained in books, because you really want opt-out rather than opt-in?" Compelling — and structurally identical to the case for entering China, which held that constrained information access beats none. Both arguments may be right. What they share is a premise Google never examined: that Google was the right entity to make the call.

The founding mythology excluded that question from the start. Brin and Page had both attended Montessori schools, where children direct their own learning and adults offer guidance rather than authority; they designed Google on the same principle. They spent years refusing "adult supervision," ran their IPO as a deliberate insult to Wall Street's gatekeepers, and built their corporate motto out of contempt for companies that needed laminated cards to remember not to cheat. The Montessori instinct ran through every major decision they made.

"Don't be evil" worked when the people who understood what it meant could fit in a single room. It couldn't carry the weight of decisions about global book archives, surveillance infrastructure, and what a billion people in China were permitted to find.

When Eric Schmidt handed the CEO role back to Page in April 2011, he tweeted: "Day-to-day adult supervision no longer required." A joke. But the joke had always been the governance model — and at this scale, that was the problem.

The Cow Suit Was Not a Metaphor — Until It Was

The cow suit at Halloween 1999 wasn't a stunt. Brin and Page genuinely didn't care how they appeared, because the work was the point. That indifference — to pageview metrics, to Wall Street gatekeepers, to what advertising was supposed to be — produced everything worth admiring in this story. It also produced every failure. The same quality that let them ignore stickiness so search could actually work let them ignore a domestic violence survivor's exposed contacts, a dissident's Gmail, a publisher's right to refuse. Indifference to convention and indifference to harm are hard to tell apart from the inside. The question the book leaves you with isn't whether Google became evil. It's whether a company that organizes the world's information can define what "the right thing" means when users, governments, publishers, and the open web all need different answers — with no external check to settle it.