Work Rules!

By Laszlo Bock

11 min read

Why does it matter? Because the management systems most companies use are perfectly designed to drive away their best people.

Most managers, if they're being honest, don't actually believe people are good. Not at a gut level. They believe people need structure, oversight, incentives, consequences — that without these guardrails, things fall apart. Every bureaucracy ever built is a monument to that suspicion. And here's the uncomfortable part: those managers aren't wrong about human nature. They're wrong about the math. What Bock found after a decade of running controlled experiments at Google — stripping away the traditional machinery of management, the ratings, the forced rankings, the boss who decides your fate, and replacing it with radical transparency and genuine freedom — doesn't just challenge the philosophy of control. It dismantles it. Work Rules! isn't a utopian vision. It's a measurable argument, and by the end, the harder question isn't whether it works. It's why nobody else is doing it.

The Control Assumption Is Costing You Your Best People

Control feels safer than freedom. When you can set the targets, monitor the output, and fire the underperformers, you at least know where you stand. The problem is that this logic is costing you roughly half your productivity — and that's not a soft claim about culture or morale. It's a manufacturing number.

MIT researcher Richard Locke studied two Nike T-shirt factories in Mexico doing identical work under the same corporate parent. At Plant B, managers owned every decision: tasks were assigned, rules were strict, the shop floor was tightly supervised. At Plant A, workers helped set their own production targets, organized themselves into teams, and had authority to halt the line when they spotted problems. Plant A produced 150 shirts per day at eleven cents each. Plant B produced 80 shirts at eighteen cents each. Same product, same country, same company — the empowered plant was nearly twice as productive at nearly half the cost per unit.

Kamal Birdi at the University of Sheffield tracked 308 companies over twenty-two years. Every company had invested in standard efficiency programs — and across all 308 firms, those programs produced no consistent improvement in aggregate productivity. Zero. The only changes that reliably moved the needle were giving employees real decision-making authority, broadening learning beyond job-specific skills, and increasing team autonomy. Together, those factors added nine percent more value per employee. In a low-margin, high-volume business, nine percent per employee isn't a rounding error. It's the difference between a plant that survives and one that doesn't.

The oversight structures most organizations treat as basic competence are not neutral. They are actively suppressing output. You can't afford more freedom is the wrong question. The right one is how much the alternative is already costing you.

Hiring Is the One Decision That Compounds Forever — Most Companies Treat It Like an Afterthought

What if the thing you can't fix undoes everything you spend on fixing people?

Most organizations treat hiring and training as interchangeable levers — if you can't find great people, you develop the ones you have. The math seems to work. Except it doesn't. One meta-analysis of corporate training found that roughly 90 percent of it produces no sustained change in how people actually do their jobs. You can run someone through a leadership program, a communication workshop, a technical bootcamp — and within weeks, behavior reverts. The classroom version of a skill and the practiced version are almost entirely different things. Training has its place, but betting your organization's quality on it is roughly as reliable as betting on the weather.

Hiring is the opposite kind of decision. You can fire someone who doesn't work out, but you can't easily upgrade them. The ceiling of what a person can become in your organization is set, in large part, the day they walk in. Alan Eustace, a senior vice president at Google, put a number on it: a genuinely exceptional engineer is worth roughly 300 times an average one. If that's even directionally right, then one hiring decision at the top of the distribution outweighs years of training programs aimed at the middle.

Yet the average company spends about $600 per employee on training and $450 on hiring. Google inverts this entirely, spending more than twice the industry average on recruiting. That's not because Google is indifferent to developing people. It's because they understand which investment is reversible. A training program that doesn't work can be cancelled next quarter. A mediocre hire compounds forward — into their own output, into the norms they model for colleagues, into the hiring decisions they'll eventually make themselves.

The implication is uncomfortable: the organizations most convinced they have a talent development problem probably have a talent selection problem they haven't looked at yet.

Your Gut Is a Terrible Interviewer

Picture yourself midway through a promising interview. The candidate is sharp, well-prepared, and you've already decided they're the one — maybe five minutes in. The rest of the conversation feels like confirmation. You leave the room confident you can read people. You probably can't.

In 1998, psychologists Frank Schmidt and John Hunter compiled eighty-five years of research on hiring and performance. They wanted to know which assessment methods actually predicted how someone would do the job. The results made conventional hiring look embarrassing. Unstructured interviews — the kind where a manager asks whatever comes to mind and trusts their instincts — explained only 14 percent of subsequent job performance. That's slightly better than checking someone's references (7 percent) or counting their years of experience (3 percent). It is nowhere near predictive.

The reason isn't that interviewers are foolish. It's that the brain has already done its work by the time the questions begin. Most interviewers arrive at a judgment within the first ten seconds of meeting someone. The rest of the conversation is spent finding evidence for that verdict. The first impression reliably predicts the interviewer's final hire/no-hire decision. It does not reliably predict how the candidate will actually perform once on the job. Those two things — who we'd like to hire and who will do the work well — turn out to be almost unrelated.

The Schmidt and Hunter data also pointed toward what does work. Work sample tests, where candidates complete a task similar to the actual job, predicted 29 percent of performance. Structured interviews — where every candidate answers the same questions, scored against defined criteria — predicted 26 percent. Both approaches require more upfront effort than a conversational interview. Neither requires a bigger budget. They require only the willingness to trade the comfortable illusion of good judgment for a method that's actually accurate.

At Google, this translated into a four-attribute model: general cognitive ability (the capacity to learn), emergent leadership, intellectual humility, and role-specific knowledge — listed last deliberately, because deep expertise often means reaching for familiar solutions when the problem demands a fresh one. The most interesting attribute is emergent leadership, which Google defined as the ability to step into a problem and, just as importantly, step back out — handing off ownership once your contribution is done rather than clinging to the role. Your gut would rank them in reverse.

Stripping Managers of Power Is How You Get Better Managers

Stanley Milgram ran a simple experiment at Yale in the 1960s. Participants were told to administer electric shocks to a stranger every time the stranger answered a question incorrectly. The shocks climbed in voltage, the stranger screamed, then went silent. Sixty-five percent of participants kept going all the way to 450 volts — what the dial labeled "Danger: Severe Shock" — not because they wanted to hurt anyone, but because someone in a lab coat told them to. Not one person stopped without first asking permission. The lesson isn't that people are cruel. It's that authority has a gravitational pull strong enough to override ordinary conscience.

Google's response to this wasn't a training program or a values poster. It was structural. Managers at Google cannot unilaterally hire, fire, rate performance, set pay, or approve promotions. Every one of those decisions runs through committees or peer groups with no direct stake in the outcome. New managers typically resist this — it feels like accountability without tools. Then they notice what changes: their relationships with their teams shift when employees know their livelihood doesn't hinge on keeping one person happy. The fairness isn't just real; it's verifiable. Google's internal research team ran the numbers and found promotion rates were consistent regardless of which project you worked on or whose desk you sat near. That kind of transparency is only possible when no single person controls the outcome. Strip the power, and what's left — mission, judgment, coaching — turns out to be what good management actually was all along.

Most Performance Reviews Guarantee the Conversation You're Trying to Have Won't Happen

Don walked into his year-end performance review having already won. For the previous three months, he had quietly negotiated every expectation downward — resetting targets, softening language, buying himself room — so that when the ratings came out, his numbers looked strong. He wasn't being dishonest exactly. He was being rational. The system rewarded him for gaming it, so he gamed it.

That's what happens when you ask one conversation to do two incompatible things. A performance review is supposed to help someone grow. A pay discussion is supposed to distribute limited money based on where you rank. The moment both happen in the same meeting, the employee stops thinking about growth and starts calculating. And they're right to. If a number is about to be attached to your name that determines your raise, your bonus, and how your manager sees you for the next twelve months, learning is not what your brain is optimizing for.

In a classic lab experiment, researchers gave people three-dimensional puzzles to solve. In the first session, everyone just worked. In the second, one group was paid per puzzle — and their engagement jumped 26 percent. In the third session, the money ran out. Time spent on the puzzles didn't drop back to baseline. It dropped 37 percent below where it started, before anyone had ever been paid. Introducing an external reward had permanently changed how people thought about the work.

Google's solution looks almost embarrassingly simple once you've seen what breaks without it: never have those two conversations at the same time. Laszlo Bock pushed to stagger them after watching too many review cycles collapse into salary negotiation. Annual reviews happen in November. Pay decisions come a month later. Stock grants are settled six months after that. Each conversation gets to be what it needs to be, because the money isn't on the table yet. When the learning brain and the calculating brain aren't competing for the same meeting, the learning brain actually shows up.

Your Best Performers Are Worth Treating Completely Differently — and So Are Your Worst

Pay everyone in the same role roughly the same amount and you will congratulate yourself on fairness while systematically subsidizing mediocrity with money that should go to your best people. That's the logical consequence of treating performance as a bell curve, where most people cluster near average and only a few stray to the extremes. But performance in most jobs doesn't follow a bell curve — it follows a power law: a small group of exceptional contributors doesn't outperform the average by ten or twenty percent, they outperform by orders of magnitude. When the distribution is that skewed, equal pay isn't equitable. It's a transfer payment.

The clearest illustration is the stock award gap Google documented among employees doing nominally identical work. Two people, same role, same level — one received a stock award of ten thousand dollars, the other one million. That's a hundred-to-one spread within a single job classification. The underlying logic extends throughout the system: pay variance of three hundred to five hundred percent at the same level is standard practice at Google, not an exception. The justification isn't favoritism. It's that the process producing the number is transparent and consistent, so employees can understand exactly why the gap exists and what it would take to close it. Procedural fairness — trust that the method is honest — turns out to be what people actually need to accept a result that might otherwise feel insulting.

The same logic runs in the opposite direction with underperformers. Rather than running annual culls, Google identifies the bottom five percent and opens with a disarmingly direct message: you're in this group, and we want to help you improve. The question being diagnosed isn't whether to fire someone — it's whether the gap is a skills problem or a fit problem. Employees moved through this process often land near the fiftieth percentile afterward. Moving one person from the fifth percentile to the fiftieth costs less than recruiting a replacement and produces more.

The Most Powerful Benefits Cost Almost Nothing

Every six months, a Googler with a history of cancer lay inside a medical scanner and waited. And every six months, while waiting for results that could go either way, he mentally drafted an email to Larry Page asking that his unvested stock continue flowing to his family after he was gone. He wrote and rewrote this email in his head on the scanner table, then went home, and wrote it again six months later. That changed the day Google quietly introduced a new death benefit: if a Googler died, their partner would immediately receive all unvested stock, plus fifty percent of the employee's salary for ten years, plus a thousand dollars a month per child. It was just the right thing to do. When the benefit eventually became public, the first question Bock heard from peers at other companies was predictable: doesn't that cost a fortune? The actual number was 0.1 percent of payroll. The cancer survivor sent Bock a note saying he'd had a scan two weeks after hearing the news. He wrote no emails in his head.

This is what most budget conversations miss entirely. Organizations spend lavishly on the visible layer of culture — catered food, commuter shuttles, gym equipment — and then treat the moments of genuine vulnerability as line items to minimize. But the math runs the other way. Extending maternity leave from twelve to twenty weeks at full pay sounds expensive until you track what was happening before: new mothers were leaving at twice the rate of other employees. Fixing that spike in attrition eliminated a cost that was already there, hidden in recruiting and onboarding. The extended leave paid for itself. A five-hundred-dollar cash gift to new parents for takeout meals costs almost nothing and signals something the daily free lunch cannot: that the company sees you at 2 a.m. with a newborn, not just at noon in the cafeteria. The perks employees talk about most are rarely the expensive ones. They're the ones that showed up when the employee was most exposed.

You Can Nudge People Toward Better Decisions Without Removing Any Choices

What if you could change the way thousands of people eat, save, and work — without mandating anything, without spending much, and without anyone noticing you'd intervened? That sounds like a trick question. It isn't.

Google tested this in its New York office by moving candy into opaque containers and putting fresh fruit and vegetables into the glass jars where the sweets used to sit. Nobody was told what to eat. The vending machines still worked. But in seven weeks, 3.1 million fewer calories were consumed — roughly 885 pounds of weight gain that simply never happened. The options didn't change. The visibility did.

The same logic applies to the first day of work. New hires' ramp-up time varied enormously depending on how their managers handled the first week. The response wasn't a training program. It was a single email, sent to managers on the Sunday evening before a new hire arrived, listing five almost embarrassingly simple actions: have a conversation about roles and expectations, pair the new hire with a peer buddy, help them build a network, schedule monthly check-ins for the first six months, encourage honest dialogue. Managers who acted on the email got new hires to full productivity 25 percent faster — a full month of learning time recovered. One email, no new budget, no new policy.

The through-line in both cases is the same: people's behavior is already being shaped by what surrounds them. The candy is visible or it isn't. The email arrives Sunday or it doesn't. The environment is always shaping behavior — the only question is whether you do it on purpose. Deliberate is almost always cheaper.

High Trust Is Real and Messy — and the Mess Is the Point

In April 2008, a chef at a Google cafeteria made a pie. He named it after its ingredients — goji berries sourced from Tibet — calling it the Free Tibet Goji-Chocolate Crème Pie. Within hours, a Googler had emailed CEO Eric Schmidt threatening to quit over what they read as a political statement about Chinese sovereignty. The email got forwarded to a company-wide mailing list. Then came 1,300 more emails. Mainland Chinese Googlers argued the name was an affront. Others insisted a chef should be able to name a dessert whatever he liked. Someone suggested the logical extensions: Free Wales Pie, War of Northern Aggression Hotcakes. The chef was sent home on a three-day suspension, which sparked a separate argument about whether that would chill speech across the company. Eventually the thread slowed and died. Nothing was resolved. Nobody changed their mind.

Bock's conclusion, reading through hundreds of those emails, was not that the debate had gotten out of hand. It was that the debate was proof the culture worked. When you genuinely give people voice and transparency, you do not get an organization of cheerful consensus. You get enormous, brawling, inconclusive arguments about pie. You also get a Googler threatening to defect to Facebook over a meatless Monday. You get roughly one major leak per year — and a termination the morning after each one, which is the price Google consciously pays to keep everything else open.

The high-trust model earns credibility precisely because of these embarrassments, not despite them. A workplace where 1,300 emails fly over a dessert name is a workplace where people believe their voices count for something. That's not a polished outcome. It's the real one.

The Question Underneath All Ten Rules

Bock's ten rules come down to a single question you have to answer before any of them make sense: do you actually believe people are fundamentally good? Not as a sentiment you'd endorse in a meeting, but as a load-bearing assumption. Most management systems were built by people who quietly answered no, then rebranded that answer as accountability and rigor. The discomfort of doing this differently — sharing information that feels dangerous to share, tolerating the 1,300-email argument about a pie, paying in ways that look unfair until you explain the logic — isn't incidental to the model. It's the evidence that you mean it. What matters isn't whether Google's rules work for your organization. It's whether you're willing to find out what happens when you stop designing for the exception and start building for the person you actually hired. That willingness is the load-bearing assumption — the thing your policies reveal when nobody's watching.