What the Productivity Books Get Wrong — and What's Left

In 2011, Daniel Kahneman wrote about a strange set of experiments. People who read words about old age walked more slowly down a corridor afterwards. People shown pictures of money became more selfish. Subtle cues, the studies said, steer behaviour without our noticing. Kahneman, a Nobel laureate and one of the most careful thinkers in psychology, was unusually firm: "disbelief is not an option."

A few years later, disbelief became the only option. The effects failed to replicate, and when researchers examined the articles cited in that chapter, the evidence collapsed. Kahneman, to his great credit, said so in public.

If that can happen to Thinking, Fast and Slow, it can happen to any book on your shelf. So this essay does something slightly tedious and, it turns out, quite revealing. It takes 23 popular books on productivity, from Atomic Habits and Deep Work to The Mythical Man-Month and Why We Sleep, breaks them into their individual claims, and checks each claim against the research.

How the books were scored

Each claim got a rough number: the estimated chance that the claim, as the book states it, holds up against the evidence today. The scale has seven steps:

• ~95%: mathematics, or many large replications. You can build on it.
• ~85%: strong evidence with minor caveats about size or generality.
• ~70%: the direction is right, but the book overstates how big, how general or how causal it is.
• ~50%: genuinely open, mixed, or plausible but untested.
• ~30%: weak, or partly contradicted.
• ~15%: mostly contradicted, or a key study failed to replicate.
• ~5%: failed large preregistered replications, retracted, or conceded by the author.

These numbers appear in brackets after the key claims in each chapter, like this: (~85%). They are judgement calls, not measurements. Where they come from is the evidence, not the author's charisma: meta-analyses where they exist, then large replications, then single studies. Each time, the question is who was studied (students, call-centre staff, sixteen programmers) and whether the result was ever repeated. Where possible, the original papers were read, not the books' summaries of them. That habit alone changed several verdicts.

What came out

Of the 377 claims, about a quarter rate ~85% or higher. About a third sit at ~70%. Roughly one in four is a genuine coin flip, and about one in seven is weak or wrong.

That middle band is the real story. The single most common verdict is "right direction, wrong size". The books rarely invent a mechanism out of nothing. They take a real effect and inflate it, generalise it from students to everyone, or sell a correlation as a cause. Planning really does help you follow through, but the best current estimate of the effect is between a quarter and a half of what the books quote. Habits really do form through repetition, but "66 days" is the median of a wide spread from a small study.

The second pattern is less comfortable. The more motivational the advice, the weaker the science behind it. The idea that willpower is a tank that runs dry was tested in two multi-lab replications with more than 5,000 participants and found essentially nothing. "Get 1% better every day and you'll be 37 times better in a year" is correct arithmetic and an incorrect description of how anyone improves. These are the claims that make books sell, and they rate ~5% to ~30%.

Meanwhile, the most reliable material in the pile is the most boring: queues, bottlenecks, estimation errors, the cost of coordinating people. Much of it is not psychology at all but mathematics, which has the pleasant property of not failing to replicate. Two of the three highest-scoring books in this collection are The Goal and The Phoenix Project. Both are novels. The most trustworthy productivity advice on the shelf comes with fictional characters and, in one case, a plant manager whose marriage is falling apart.

Averages flatter short books and punish ambitious ones, so this isn't a league table. Deep Work scores lowest on average partly because it makes so many bold, checkable claims. But the pattern holds across the collection: where the books lean on operations research and judgement research, they're solid. Where they lean on brain metaphors and social psychology from before the replication crisis, they wobble.

What "productive" means here

Productivity is output divided by input. Both halves are routinely mismeasured.

Output is valuable results, not activity. Emails sent, hours logged and lines of code written are activity. A finished project, a solved problem, a patient discharged on time are output. Most of the bad advice in this field comes from optimising the wrong one.

Input is not just hours. It is time, attention and energy, and those three don't trade one for one. An hour of fragmented attention after a short night is a different input from an hour of focus after a good one, even though a timesheet can't tell them apart.

That gives two ways to improve: get more valuable output from the same input, or get the same output with less. The second is underrated. Nobody writes bestsellers called Achieve the Same With Less Suffering, but much of what survives the evidence is exactly that.

A production line with six stations

What's left after the checking fits a simple picture: your work as a small production line. Raw material (ideas, requests, obligations) comes in. Finished, valuable results go out. In between, six stations decide how much comes out and at what cost. Each chapter covers one.

1. Choose. Are you working on the right things, and is the decision made before the moment of temptation? Goals, plans, decisions, motivation.
2. Flow. Does work move through the line, or pile up in front of it? Work in progress, slack, bottlenecks, estimation.
3. Attention. Does the work get undivided attention when it needs it? Focus, interruptions, switching.
4. Energy. Is the machine maintained? Sleep, hours, breaks, recovery.
5. Leverage. What multiplies output: skill, tools, AI, other people?
6. Feedback. How do you know any of it works, without fooling yourself?

The order isn't arbitrary. Choosing comes first because perfect flow on the wrong work is just efficient waste. Feedback comes last because it closes the loop and sends you back to choosing.

Each chapter follows the same pattern: what the books claim, what the evidence shows, and why the surviving part works. Two boxes end each chapter. What doesn't work collects the claims that failed, which are often the most useful part. Try this lists practices that follow from the evidence rather than from the authors' enthusiasm.

A word of caution before starting. This is a careful reading, not a meta-analysis. The ratings are estimates, the selection of books is one person's shelf, and the science will move on. Some of these verdicts will be wrong. The aim is not to be the final word, but to be honest about which words are load-bearing.

1. Choose: most decisions are made before the moment

At lunchtime in the cafeteria of Massachusetts General Hospital, people reach into the fridges and pick a drink. In 2012 a team of researchers changed what came out of those fridges without asking anyone to try harder. First they put traffic-light labels on everything, and sales of red-labelled drinks, mostly sugary ones, fell 16.5%. Then they rearranged the shelves so that water and healthier items were easier to see and reach. Red drinks fell a further 11.4% and bottled water rose 25.8%, while two comparison cafeterias saw no such shift (Thorndike et al. 2012). Two years later, red drinks were still down, from 26% to 17% of drink sales (Thorndike et al. 2014).

The person at the fridge feels as if they are choosing. Most of the choosing was done earlier, by whoever stocked the shelves. Work is like that too. By Monday morning, the important decisions (which projects you said yes to, what you're aiming for, what happens when the work gets unpleasant) have mostly been made, deliberately or by default. Peter Drucker put the stakes plainly: nothing is so useless as doing with great efficiency what should not be done at all (Drucker 1963).

So the first station asks two things. Am I working on the right things? And is the decision made in advance, rather than at the fridge door?

Specific goals, unless you're still learning

Atomic Habits tells readers to forget about goals and build systems instead, because winners and losers share the same goals. Winners and losers also share alarm clocks. The research on goal-setting disagrees.

Edwin Locke and Gary Latham's summary of the field found that specific, hard goals beat "do your best" by a moderate to large margin (d = .42 to .80), across well over 100 tasks, more than 40,000 participants and at least eight countries (Locke & Latham 2002) (~85%). The caveat is that the summary was written by the theory's founders. The logic is easy to see: "do your best" draws no line to cross, so almost any effort can count as your best.

There is an exception. The effect shrinks as tasks get more complex: d = .41 for the most complex against .77 for the simplest (Wood, Mento & Locke 1987). In an air-traffic-control simulation, an outcome goal actually hurt learning (Kanfer & Ackerman 1989); a learning goal, such as "find three strategies that work", fixed it (Winters & Latham 1996) (~70%). A nurse in her first week on a new ward doesn't need a target for patients per shift. She needs a target for what to figure out.

Clear's "systems" aren't worthless, though. Much of what he means overlaps with the next tool.

Fantasise, then find the obstacle

Gabriele Oettingen's Rethinking Positive Thinking sums up two decades of her lab's work on imagining success. In one set of studies, graduates who fantasised more about a great job sent out fewer applications, got fewer offers and earned lower salaries two years later (Oettingen & Mayer 2002). Expectations based on experience had the opposite effect. The samples were small and come from one lab, but the distinction is useful: confidence helps, while daydreaming feels like progress without being any.

Her fix, mental contrasting, pairs the wish with the main inner obstacle: a habit, a mood, a belief. Add an if-then plan and you get WOOP: wish, outcome, obstacle, plan. A meta-analysis of 21 studies found a small-to-moderate benefit (g = 0.34), possibly inflated by publication bias; versions people ran on their own did worse than ones guided by a researcher (Wang, Wang & Gai 2021).

If-then plans ("if it's 4 pm and the report isn't started, then I close my email and open the draft") are the best-tested part. The figure that went into many books was d = .65, from a 2006 meta-analysis of 94 tests (Gollwitzer & Sheeran 2006). The 2024 update, with 642 tests, found d = .36, and about .15 after correcting for publication bias (Sheeran, Listrom & Gollwitzer 2024) (~85%). The authors call the effect robust but small. It was larger with strict if-then wording (.43 against .29 for schedule-style plans), high motivation and rehearsal. An if-then plan is cheap rather than powerful, which is a fine thing for a tool to be.

What should the plan target? Piers Steel's meta-analysis of 691 correlations gives an answer (Steel 2007) (~85%). People put off tasks that are unpleasant (r = .40), and the people who put things off tend to be impulsive (.41), distractible (.45) and low in conscientiousness (−.62). Perfectionism, the most flattering self-diagnosis available, barely registers (−.03). Procrastination also costs performance (r = −.19). The data describe who procrastinates, not what cures it, but they point the plan at the right moment: the unpleasant start and the tempting distraction.

Habits: decide once, then stop deciding

A habit is a decision you no longer have to make. The question is how long the handover takes. In the best-known study, 96 volunteers each picked a new daily behaviour; for the 39 whose data fitted a curve well, it took a median of 66 days to get 95% of the way to automatic, with a range of 18 to 254 days (Lally et al. 2010) (~85%). The upper end was projected beyond the 84-day study, and automaticity was self-reported. Missing a day or two made no significant difference.

The cafeteria shows the other half: the environment does much of the deciding. In a study of 205 adults reporting 7,827 desires, people high in self-control mostly avoided temptations rather than resisting them (Hofmann et al. 2012), though the data are correlational. Making the good option easier and the bad one harder is well supported in direction; the size varies a lot by setting.

Temptation bundling, pairing something you want with something you should do, shows a common pattern. In the first experiment, people who could only listen to gripping audiobooks at the gym went 51% more often, though the effect faded, especially after Thanksgiving (Milkman, Minson & Volpp 2014). A follow-up with 6,792 people raised the chance of a weekly workout by 10–14% (Kirgios et al. 2020). Big in the first small study, smaller but real at scale. (The famous "1% better every day" is a separate problem; see Feedback.)

Simple rules beat gut feel

Some decisions come up again and again: whom to hire, which requests to accept, which patient to see first. Here the evidence is about as strong as psychology gets. Across 136 studies, simple statistical rules were about 10% more accurate than expert judgment on average; experts did clearly better in only 6–16% of studies, whatever the task or their experience (Grove et al. 2000) (~95%).

Hiring is the everyday version. A structured interview, with the same questions for every candidate and each answer scored on a fixed scale, predicts job performance with a validity of about .42; a free-flowing conversation manages about .19 (Sackett et al. 2022). Drucker made the general point: if a problem is generic, solve it once with a rule instead of treating every case as unique (Drucker 1967).

For one-off decisions there is the premortem: before committing, the team imagines the project has failed a year from now and writes down why. In the study behind it, imagining the outcome as already happened raised the number of reasons people found by about 30% (Mitchell, Russo & Pennington 1989). It lets the doubters speak before the money is spent.

Afterwards, judge the decision by how it was made, not only by how it turned out; why that is harder than it sounds is a question for Feedback.

Motivation: rewards, autonomy and passion

Daniel Pink's Drive argues that carrots and sticks backfire on creative work and that people are really driven by autonomy, mastery and purpose. The core holds. Across 128 experiments, expected tangible rewards reduced people's interest in a task for its own sake once the reward stopped (d = −.28 to −.40; Deci, Koestner & Ryan 1999) (~85%). Children who expected a "Good Player" award for drawing drew less afterwards than children who got the same award unexpectedly (Lepper, Greene & Nisbett 1973). Rewards can also change what a behaviour means: when Israeli daycares fined parents for late pickup, lateness went up and stayed up after the fine was removed (Gneezy & Rustichini 2000). The fine had become a price.

The headline claim, that rewards make people worse at creative work, is where the book overreaches. Its showcase, the candle problem, found that offering money slowed people down on a puzzle that needs insight (Glucksberg 1962); a 2013 replication found no negative effect (Ramm, Tjøtta & Torsvik 2013). A later meta-analysis found incentives helped on interesting tasks about as much as on dull ones (δ ≈ .58 against .52), though less for quality (Kim, Gerhart & Fang 2022). The better summary is "both": across 183 studies, intrinsic motivation predicted the quality of work, incentives predicted quantity, and the two worked at the same time (Cerasoli, Nicklin & Ford 2014) (~70%). Pink is directionally right and about one size too strong.

Cal Newport's So Good They Can't Ignore You asks where the love of the work comes from. He argues that "follow your passion" is bad advice, because most people don't have a passion that maps onto a job. Of 539 Québec college students, 84% had an activity they felt passionate about, but only 3.6% of those activities had to do with work or study; about 60% were sports (Vallerand et al. 2003). Interests also seem to grow rather than wait to be found: in short studies, mostly with students, those who believed interests are fixed lost interest faster when a new topic got hard (O'Keefe, Dweck & Walton 2018).

The strong version, that following your passion is the worse strategy, isn't supported. People who look for a fit and people who expect to develop their passion end up with similar passion, satisfaction and income, by different routes (Chen, Ellsworth & Schwarz 2015) (~30%). Newport's alternative, building rare and valuable skill, belongs to Leverage.

Choosing what to neglect

The last decision is the least comfortable. An average life lasts about 4,000 weeks, roughly 77 years. That's arithmetic, and its consequence is logic: every yes is a no to everything else that could have filled the same hours. Oliver Burkeman's Four Thousand Weeks argues that the skill isn't doing more but choosing well what to fail at (Burkeman 2021). His claim that efficiency mostly brings more work, the way a new road fills with traffic, is plausible but untested.

Drucker had a word for the practical side: posteriorities. Choosing priorities is easy; deciding what to stop doing is harder and matters more. His test is to ask of every activity, "What would happen if this were not done at all?", and drop whatever gets the answer "nothing" (Drucker 1967) (~70%). That rating rests on a mechanism, not a test of the question: fewer commitments in parallel means each finishes sooner (see Flow). The popular version, Warren Buffett's "two lists" of 25 goals with 20 to avoid at all costs, comes second-hand from his pilot and was never confirmed by Buffett. The idea doesn't need him.

Like the cafeteria shelves, a list of commitments is stocked long before anyone stands in front of it. Most productivity advice is about getting through the list faster. This station is about what goes on it.

What doesn't work

• "Forget goals, build systems." Specific, hard goals beat "do your best" by d = .42–.80 across more than 100 tasks (Locke & Latham 2002). The book's claim: ~30%.
• "Rewards kill creativity." The candle problem didn't replicate, and incentives help on interesting tasks too (Ramm et al. 2013; Kim et al. 2022). ~15%.
• "I procrastinate because I'm a perfectionist." Across 691 correlations, the link is about zero (r = −.03); unpleasant tasks and impulsiveness matter far more (Steel 2007). Steel's pattern: ~85%.
• "Mental contrasting rewires you unconsciously." A new-data replication found no difference from the control condition (Kappes et al. 2012). ~15%.
• "Sleep on it and your unconscious will decide better." The unconscious-thought effect failed a 2015 meta-analysis and replication (Nieuwenstein et al. 2015). ~15%. A plain pause before a hot decision still helps (~70%).

Try this

• Turn "do your best" into a number and a date. If the task is new to you, start with a learning goal instead: "find three ways to…".
• Run a five-minute WOOP for the week's most important task. Name the inner obstacle, then write the plan in strict form: "If [obstacle], then I will [action]." Rehearse it once.
• Move the shelves. For one habit, put the good option within reach and add a step to the bad one. Expect weeks to months, not days, and don't treat a missed day as a reset.
• Write down a rule for one recurring decision, such as which requests you accept or how you score candidates. Before any large project, hold a premortem.
• Once a quarter, list every commitment and ask Drucker's question: what would happen if this were not done at all? Drop at least one.

Sources

• Burkeman 2021, Four Thousand Weeks
• Cerasoli, Nicklin & Ford 2014, Intrinsic motivation and extrinsic incentives jointly predict performance
• Chen, Ellsworth & Schwarz 2015, Finding a fit or developing it
• Clear 2018, Atomic Habits
• Deci, Koestner & Ryan 1999, meta-analysis of extrinsic rewards and intrinsic motivation
• Drucker 1963, Managing for business effectiveness (Harvard Business Review)
• Drucker 1967, The Effective Executive
• Glucksberg 1962, the candle problem and incentives
• Gneezy & Rustichini 2000, A fine is a price
• Gollwitzer & Sheeran 2006, implementation intentions meta-analysis (94 tests)
• Grove et al. 2000, Clinical versus mechanical prediction: a meta-analysis
• Hofmann et al. 2012, Everyday temptations
• Kanfer & Ackerman 1989, goals and skill acquisition in an air-traffic simulation
• Kappes et al. 2012, mental contrasting and obstacle–behaviour links
• Kim, Gerhart & Fang 2022, incentives and task interest (meta-analysis)
• Kirgios et al. 2020, temptation bundling at scale
• Lally et al. 2010, How are habits formed
• Lepper, Greene & Nisbett 1973, Undermining children's intrinsic interest with extrinsic reward
• Locke & Latham 2002, Building a practically useful theory of goal setting and task motivation
• Milkman, Minson & Volpp 2014, Holding the Hunger Games hostage at the gym
• Mitchell, Russo & Pennington 1989, Back to the future: temporal perspective in the explanation of events
• Newport 2012, So Good They Can't Ignore You
• Nieuwenstein et al. 2015, unconscious thought (meta-analysis and replication)
• O'Keefe, Dweck & Walton 2018, Implicit theories of interest
• Oettingen 2014, Rethinking Positive Thinking
• Oettingen & Mayer 2002, The motivating function of thinking about the future
• Pink 2009, Drive
• Ramm, Tjøtta & Torsvik 2013, incentives and the candle problem (CESifo WP 4374)
• Sackett et al. 2022, validity of selection methods
• Sheeran, Listrom & Gollwitzer 2024, implementation intentions meta-analysis (642 tests)
• Steel 2007, The nature of procrastination
• Thorndike et al. 2012, labelling and choice architecture in a hospital cafeteria
• Thorndike et al. 2014, cafeteria choice architecture at 24 months
• Vallerand et al. 2003, Les passions de l'âme
• Wang, Wang & Gai 2021, MCII meta-analysis
• Winters & Latham 1996, learning goals versus outcome goals
• Wood, Mento & Locke 1987, task complexity and goal effects

2. Flow: the arithmetic of the pile

Picture a supermarket on a Saturday afternoon. Every till is staffed, every cashier is scanning as fast as they can, and the queues still reach back into the frozen aisle. Nobody is lazy. The shop is simply running at full capacity, and a shop at full capacity is a shop where customers wait.

Most knowledge workers run their week like that supermarket. Every hour is booked, a dozen projects are "in progress", and the important thing still takes a month. The usual diagnosis is effort or discipline. The usual cure is a better to-do app.

This chapter is about a different diagnosis. Work behaves like a queue, and queues follow rules that have nothing to do with willpower. Several of the rules are mathematics, which makes this the rare part of the productivity literature where the books are mostly right. The question for this station is plain: does work move, or does it pile up?

Little's law: why more in progress means slower

In 1961 John Little proved a result so general it barely sounds like a finding. Average work in progress equals the arrival rate times the average time each item spends in the system. Rearranged: the time an item takes to get through equals the amount of work in progress divided by throughput. It holds for any system, over any period that starts and ends empty, however the work arrives (Little 1961; Little 2011) (~95%).

The consequence is uncomfortable. If you finish about two projects a month and have twelve open, the average project will take six months, however hard you work on each one. Starting a thirteenth doesn't add output. It adds waiting.

The Phoenix Project turns this into its central lesson: limit work in progress, because releasing more work into a busy system makes everything slower. Cal Newport's Slow Productivity makes the personal version: "do fewer things", keep a holding queue, and pull new work in only when something finishes (~85%). He adds an observation that matches everyday experience: every commitment carries an overhead tax of emails, check-ins and meetings, so the number of commitments matters more than their size. The switching-cost mechanism behind that is supported; his tipping-point picture is a model.

There is one honest caveat. Little's law is an identity of averages. Cutting work in progress shortens lead time only if throughput doesn't fall with it. Close half your projects and spend the freed time refreshing your inbox, and nothing improves.

The field evidence for the practical tools is thinner than the maths. Reviews of kanban boards and WIP limits in software teams find the literature "largely descriptive", with the best-supported benefits being visibility and flow; one four-year case study of five teams found that lower WIP went with shorter lead times, but nobody knows the optimal limit.

Small batches follow the same logic. A big batch is a lot of work in progress travelling together, and nothing in it is finished until all of it is. In one production-line simulation, cutting batch sizes reduced lead time by around 30%, though extra setups can eat the gain. In software, the State of DevOps surveys found that teams deploying often, in small pieces, also had fewer failed changes. The direction has held across years; the dramatic multipliers ("46× more deployments") depend on how the survey answers were bucketed, so treat them as decoration (direction ~70%, multipliers ~30%). (The metrics themselves: see Feedback.)

Slack: why full is broken

Back to the supermarket. Queueing theory explains why the queue grows so suddenly in the last stretch before full capacity. In the simplest queue, waiting time scales with utilisation divided by idle time. At 50% busy, that ratio is 1. At 90% busy, it is 9: work waits nine times as long (~85%). Push towards 100% and the wait heads off towards infinity.

The exact factor depends on how variable the arrivals and the tasks are. Real queues are usually more variable than the textbook one, which makes the blow-up worse, not better. Anyone who has watched a motorway at rush hour has seen this: the road isn't full, but it is full enough that one braking driver turns into a standstill.

Accelerate lists utilisation alongside lines of code as a misleading measure, because pushing it towards 100% makes queues and lead times explode. The lesson for an individual is counter-intuitive and well grounded: a calendar with no gaps is not efficient. It is a queue with nowhere for the surprises to go.

Bottlenecks: the slowest hiker sets the pace

Eliyahu Goldratt's The Goal, a 1984 business novel about a failing factory, puts the next idea in a scout troop. Herbie is the slowest boy on the hike, and however fast the others walk, the troop arrives when Herbie does. A serial system's throughput is set by its bottleneck (~95%).

Two corollaries follow, and both hold up. An hour lost at the bottleneck is an hour lost for the whole system, while an hour saved elsewhere is, for throughput at least, a mirage (savings elsewhere can still cut inventory or lead time). And keeping every worker busy is a mistake: anything produced faster than the bottleneck can absorb just becomes a pile in front of it (~85%). Local efficiency is not global productivity.

Andy Grove reached the same conclusion at Intel around the same time. His High Output Management opens with a breakfast factory and tells the reader to find the limiting step (the boiled egg) and schedule backwards from it. Goldratt's remedy is five steps: identify the constraint, exploit it, subordinate everything else to it, elevate it, and repeat when it moves. Over 80 published applications report gains in lead time and due-date performance, but they are success stories with no controls and no reported failures.

The Phoenix Project adds the bottleneck most knowledge workers will recognise: Brent, the one engineer behind almost every task, constantly interrupted and indispensable. Knowledge concentration is common. One study estimated that 65% of 133 popular GitHub projects would be in trouble if two or fewer people left. That Brent is the constraint, and that documenting his knowledge fixes it, is story. The personal version is sharper: if everything waits on you, you are your own Brent.

The same book sorts work into four types: business projects, internal projects, changes, and unplanned work. Unplanned work, the firefighting, is the dangerous one, because time is zero-sum and every emergency silently displaces something planned. Benchmark data from IT organisations suggest high performers spend under 5% of their time on it and low performers often over half (~70%; the original report isn't public). For one person, the question is the same: how much of last week was on the plan at all?

The order of work

Once work is limited and the bottleneck found, a smaller question remains: in what order? Brian Christian and Tom Griffiths's Algorithms to Live By brings in scheduling theory, and its main result is a relief. There is no best schedule, only a best schedule for a goal (~95%). Doing tasks in order of due date minimises how late the latest one is. Doing the shortest first minimises the total time things spend waiting. Moore's algorithm minimises the number of late tasks. Pick the goal before you pick the method.

Their most memorable story is priority inversion. On the Mars Pathfinder in 1997, a low-priority task held a shared resource that a high-priority task needed, and the important work stalled behind the trivial. Engineers fixed it remotely by switching on priority inheritance, which lets the blocking task borrow the high priority until it lets go. The office version is the unimportant form that blocks the important project. Sometimes the right move is to finish the trivial thing first, because something big is waiting on it.

The authors also warn about thrashing, switching so often that all time goes to switching; the cost of that belongs to the Attention chapter. And the transfer from algorithm to life is weaker than the theorems: real life rarely offers one goal, known options and no going back.

Estimates: the inside view is always sunny

All of this assumes you know how big the work is. You don't. In a 1994 study, Roger Buehler and colleagues asked 37 honours students when they would finish their thesis. They predicted 33.9 days on average and took 55.5. Only 29.7% finished by their own best guess. More striking, their worst-case estimates (48.6 days) were still too short on average, though not significantly, and fewer than half (48.7%) finished even by those (Buehler et al. 1994). These were important, familiar projects, nearly done, by people with every reason to get it right. The planning fallacy is one of the sturdiest findings in psychology (~95%).

Asking people to be pessimistic reduced the bias in that study but didn't make the estimates more accurate. The problem isn't mood. It is method. We plan from the inside, imagining how this project will go, and the imagination leaves out the flu, the missing data and the supervisor on holiday.

Daniel Kahneman's fix in Thinking, Fast and Slow is the outside view: start from how long similar projects actually took, then adjust for what's special about this one. Bent Flyvbjerg's reference class forecasting, used in British and Danish infrastructure planning, rests on the same idea; the field evidence is good, the controlled trials few (~70%). Newport's simpler rule is to double your estimates. The direction is right; the factor of two is a rule of thumb.

Frederick Brooks, writing about software in The Mythical Man-Month, adds a useful mechanism: a project gets a year late one day at a time, and fuzzy milestones hide the slippage. Milestones should be concrete enough that "done" means done, not "90% done".

More people: the coordination tax

When a project is late, the obvious move is to add people. Brooks's law says this makes it later (The Mythical Man-Month, 1975). The reasoning has two parts. Newcomers need training. And communication paths grow as n(n−1)/2: three people have three, ten people have 45 (~95%). That is an upper bound, since real teams don't use every channel, which is why structure helps.

The law itself has aged less well. Brooks later called his wording an "outrageous oversimplification". Steve McConnell argued in 1999 that it applies mainly in a project's final phases, that adding people mostly raises cost rather than always causing delay, and that most lateness comes from underestimating the project in the first place (~50% for "always later"). Coordination cost is real; the iron law is not. As Brooks put it himself, nine women can't make a baby in one month, but not every task is a pregnancy.

Conway's law describes where coordination ends up. A system's structure tends to mirror the communication structure of the organisation that builds it. In matched pairs of software products, the product built by the more loosely coupled organisation was more modular every time, by up to a factor of eight (MacCormack, Rusnak & Baldwin 2012) (~85%). A review of 142 studies found mirroring common in companies but often absent in open-source projects. The popular remedy, reorganising teams first and letting the architecture follow, is untested.

The personal lesson is about dependencies. Every task that waits for someone else's reply has a queue in it you don't control. Fewer of those is worth more than faster typing.

What doesn't work

• "Adding people to a late project always makes it later." Coordination costs are real, but McConnell and project data show the effect is mainly cost, and mainly late in a project (~50%).
• Worst-case estimates as a safety margin. In Buehler's study fewer than half of students finished by their own worst-case date (~95% for the planning fallacy).
• "Balanced plants go broke." Spare capacity away from the bottleneck is well supported, but simulations of deliberately unbalanced lines show mixed, modest gains; the slogan overstates (~70% for "keep spare capacity off the bottleneck").
• Change approval boards as the cure for outages. The diagnosis (changes cause most outages) holds; the Accelerate data found external approvals went with slower delivery and no fewer failures (~30% for the cure).
• Drum-buffer-rope as the best scheduling method. Simulation results are mixed and depend on setup and tuning (~50%).

Try this

• Set a WIP limit and keep a holding queue. Count your active projects, pick a number you can actually move each week, and start something new only when something finishes.
• Leave gaps in your calendar on purpose. Treat unbooked time as the buffer that keeps surprises from turning into a queue.
• Find your Herbie. Ask where work waits longest (your review, your decisions, someone's sign-off) and protect that step before speeding up anything else.
• Estimate from your own history. Before committing to a date, look up how long the last three similar projects took, then adjust.
• Cut dependencies. Prefer tasks you can finish without waiting for someone else, and batch the ones you can't.

Sources

• Little 1961, "A proof for the queuing formula L = λW"
• Little 2011, "Little's Law as viewed on its 50th anniversary"
• Buehler, Griffin & Ross 1994, "Exploring the 'planning fallacy'"
• MacCormack, Rusnak & Baldwin 2012, mirroring in software products
• McConnell 1999, "Brooks' Law Repealed?"
• Goldratt & Cox 1984, The Goal
• Kim, Behr & Spafford 2013, The Phoenix Project
• Forsgren, Humble & Kim 2018, Accelerate
• Brooks 1975/1995, The Mythical Man-Month
• Newport 2024, Slow Productivity
• Christian & Griffiths 2016, Algorithms to Live By
• Grove 1983, High Output Management
• Kahneman 2011, Thinking, Fast and Slow

3. Attention: the sterile cockpit

Airline pilots work under a rule most offices would find rude. In the United States, the "sterile cockpit" rule says that during the critical phases of a flight (taxiing, take-off, landing, and anything below 10,000 feet except cruise) the crew may do nothing that could distract them from flying. No small talk, no paperwork, no calls about galley supplies. Above that line, the chat can resume.

The rule doesn't assume pilots are weak-willed. It assumes that attention is limited, that some moments need all of it, and that the only reliable protection is to decide in advance what may interrupt. Nobody hands a pilot a memo about the quarterly budget during final approach.

Most offices run the other way round. Every hour is treated as cruise, open to any question, including the hours that ought to be final approach. This station asks: does the work get undivided attention when it needs it? The books are loud on this point. The research is quieter and, in places, says something different.

Blocks of time, and why small slices leak

Peter Drucker made the case in 1967 in The Effective Executive: knowledge work needs large, consolidated blocks of time, and time spent in slices of a few minutes is mostly wasted. Cal Newport turned it into a career strategy in Deep Work, with a neat formula: quality of work = time spent × intensity of focus.

Neither author tested this, and nobody has run the obvious experiment of blocks versus slices for real knowledge work. The support comes from mechanism. Switching between tasks costs time, and the cost grows with the complexity and unfamiliarity of the rules involved (Rubinstein, Meyer & Evans 2001). Preparing for a switch reduces the cost but doesn't remove it (Monsell 2003). On that basis, the claim that knowledge work needs consolidated blocks holds up reasonably well (~70%).

What a fragmented day looks like was measured in one well-known case. Leslie Perlow followed a 17-person software group and found that 75% of their uninterrupted stretches lasted an hour or less, and 60% of those half an hour or less (Perlow 1999). Nobody had designed it that way. The day had been chopped up by colleagues, one helpful question at a time.

How much deep work a day can hold is shakier. Newport's figure of about four hours for an expert comes from a study of 30 violin students, older training studies and the habits of famous writers (Ericsson et al. 1993). It has never been tested for knowledge work. Treat four hours as a generous ceiling.

Attention residue: the meeting you're still in

Switching has a second cost that outlasts the switch itself. Sophie Leroy showed in two lab experiments that when people move away from an unfinished task, part of their attention stays behind, and they do worse on the next one (Leroy 2009). She called it attention residue. Finishing the first task, even under time pressure, reduced it.

The samples were small and field data is thin, so the effect is plausible but unproven at scale (~70%). It matches a familiar experience: the email read just before a meeting that keeps replaying while someone presents the budget. The body walked into the meeting. The mind is still drafting a reply.

The practical lesson is narrower than "never switch": leaving a task unfinished costs more than leaving it finished, so the moment to go is at a natural stopping point.

Interruptions cost stress, not always speed

Here the books overreach. Deep Work treats interruptions as a drain on both time and wellbeing, and the claim travels with a famous number: it takes 23 minutes to refocus after an interruption. That number is usually credited to Gloria Mark's research. It isn't in the paper people cite.

What Mark, Gudith and Klocke (2008) found is more interesting. They gave 48 people, most of them German students, an email task and interrupted some of them. The interrupted group finished faster, with no loss of quality. They also reported more stress, frustration, time pressure and effort. People compensate for interruptions by speeding up, and pay for it in strain (~70%). The time-cost version of the claim finds no support in this study (~30%).

That shifts the argument. For short, simple tasks, interruptions may not cost output today. They cost the person producing it, and that bill arrives later (see Energy).

Perlow's software group adds a second twist. Engineers judged 96% of their interactions helpful, but only 10% urgent. The interruptions weren't junk; they were useful questions that could have waited. So the group tried quiet time: no interruptions on three mornings a week. The share of engineers rating their productivity above average went from 59% to 41% in an intermediate phase, then to 65%. The vice president credited the quiet time with an on-time product launch. Nine months later the practice had faded, and so had the effect (Perlow 1999). The finding holds for that group (~85%); whether it generalises is less certain (~50%).

Two details are worth keeping. The popular claim that quiet time "boosted productivity by 65%" misreads the result: 65% was the share of engineers reporting above-average productivity. And the effect lasted only as long as someone enforced it. Like the sterile cockpit, quiet time is a rule, not a mood.

Email behaves similarly for individuals. In one study, 124 adults spent one week limited to three email checks a day and another checking freely. Daily stress was lower in the limited week, though recalled weekly stress didn't differ (Kushlev & Dunn 2015). Fewer checks, less strain; nobody showed the work got better.

Flow: the reward for protecting the block

Mihaly Csikszentmihalyi's Flow describes what protected attention can turn into: complete absorption, with clear goals, immediate feedback, challenge matched to skill, loss of self-consciousness and an altered sense of time. People describe it in strikingly similar terms across activities and cultures. The method behind it, experience sampling, pages people at random moments to ask what they are doing and feeling, which avoids much of the distortion of memory.

Flow is more likely when challenge and skill are in balance; too easy breeds boredom, too hard breeds anxiety (Fong et al. 2015). It also goes with better performance, with a medium correlation (r = .31) in sport and gaming (Harris et al. 2021). But the authors note the direction is unclear. Playing well may produce flow as much as flow produces good play.

The most surprising finding is the "paradox of work". In a week of experience sampling with 78 workers, flow-like situations came up 54% of the time at work and 17% in leisure, yet people still said they would rather be somewhere else while working (Csikszentmihalyi & LeFevre 1989) (~85%). A later study of 100 people found the same pattern. Work, for all the complaints, is where many people do their most absorbed thinking. The default alternative is not restful: in a large smartphone sampling study, minds wandered 46.9% of the time, and people were less happy when they did (Killingsworth & Gilbert 2010).

The book's figure for the mind's processing limit, 126 bits per second, is a back-of-envelope estimate; a 2024 estimate puts conscious throughput nearer 10 bits per second. The point that attention is scarce survives. The decimal places don't.

Flow is also morally neutral. Video games and gambling produce it reliably. A state that feels this good is a reason to design work so it can happen. It says nothing about whether the activity producing it is worth doing.

Capture: getting loops out of your head

David Allen's Getting Things Done starts from a different leak. Every commitment you haven't written down, he argues, is an "open loop" that keeps tugging at your attention. His cure is to capture everything in a trusted external system and decide the next physical action for each item.

The evidence splits the idea in two. The old Zeigarnik claim that unfinished tasks are remembered better does not hold in general, but a tendency to resume unfinished tasks does (Ghibellini & Meier 2025). The pull is real; the memory version isn't.

Offloading itself is well supported. Working memory is small, about four chunks, and moving information into the world frees capacity (Risko & Gilbert 2016). In one experiment, saving a file before studying a second word list improved recall of that second list (Storm & Stone 2014) (~85%). The same study backs Allen's insistence on trust: the benefit disappeared when people were told the saving might not work. A system you don't believe in is a system you keep checking in your head.

The more specific claim, that writing a plan for an unfinished goal quiets its intrusive thoughts before you act, rests on a single paper (Masicampo & Baumeister 2011). No replication has been found either way, and it comes from the same lab and era as ego depletion (~50%). A related study fits it: 57 students who spent five minutes writing a to-do list before bed fell asleep faster than those who wrote about completed tasks (Scullin et al. 2018). One night, one lab, healthy young adults.

Two warnings from inside the system. The two-minute rule ("if it takes less than two minutes, do it now") is sound arithmetic for one task and a trap for twenty, because a chain of quick tasks is a string of interruptions you invited yourself. And as Oliver Burkeman points out in Four Thousand Weeks, a perfect inventory of everything you will never do is not the same as a clear head. Choosing what to drop belongs to an earlier station (see Choose).

Phones, multitasking and the walk outside

The books' strongest warnings about technology rest on the weakest studies. Newport claims that habitual media multitasking trains the brain to crave novelty and makes people worse at filtering out irrelevant information. The original study compared 19 heavy multitaskers with light ones (Ophir, Nass & Wagner 2009). In later replication tests, only 5 of 14 were significant, and after correcting for small-study effects the overall effect shrinks to about zero (Wiradhany & Nieuwenstein 2017) (~15%). Even if a difference existed, it could be that distractible people multitask, rather than the other way round.

The "brain drain" claim, that a smartphone merely lying nearby drains your attention, fared no better. Direct replications failed (Hartmann et al. 2021; Ruiz Pardo & Minda 2022), and a meta-analysis pooling phone use and presence found no significant effect on attention (Böttger et al. 2023) (~15%). The phone on the desk is not quietly siphoning off your concentration. Picking it up is another matter, and that is a switch like any other.

Quitting social media has modest but real effects on wellbeing. In a large experiment, four weeks off Facebook freed about an hour a day and raised a subjective-wellbeing index by 0.09 standard deviations (Allcott et al. 2020). Nobody measured what it did for work.

Breaks are where the news is better. In one experiment, 38 people did better on a backwards digit-span task after a walk of about 50 minutes in an arboretum than after one downtown (Berman, Jonides & Kaplan 2008). The samples are small. That rest restores focus is well supported; that the rest must involve trees is weaker (~70%).

What doesn't work

• "It takes 23 minutes to refocus." Not in the Mark et al. 2008 paper usually cited; that study found interrupted people worked faster, at a cost in stress. (~30% for the time-cost claim)
• "Quiet time boosted productivity by 65%." A misreading of Perlow 1999: 65% was the share of engineers rating themselves above average, and the effect vanished once the rule stopped being enforced.
• Media multitasking ruins your focus. Replications mostly failed; the bias-corrected effect is about zero (Wiradhany & Nieuwenstein 2017). (~15%)
• A phone nearby drains your brain. Direct replications failed; no significant effect on attention in the meta-analysis (Böttger et al. 2023). (~15%)

Try this

• Book one or two blocks a day in the calendar and treat them like a meeting with someone senior. Aim for well over 30 minutes, the length Perlow's engineers rarely got.
• Leave tasks at a stopping point. Before switching, finish the paragraph or the function and write down the next step, so less of you stays behind.
• Check email and phone at set times, a few times a day. Don't bother exiling the phone to another room; how often you pick it up matters more than where it lies. Expect lower stress, not miracles.
• If you run a team, agree on quiet mornings and name someone to enforce them. Perlow's version worked while it was policed and was gone nine months later.
• Keep one capture list and review it often enough to trust it, so your head stops doing the job. Between blocks, take the break outside if you can.

Sources

• Allcott, Braghieri, Eichmeyer & Gentzkow 2020, The welfare effects of social media
• Allen 2001/2015, Getting Things Done
• Berman, Jonides & Kaplan 2008, The cognitive benefits of interacting with nature
• Böttger, Poschik & Zierer 2023, smartphone use and presence (meta-analysis)
• Burkeman 2021, Four Thousand Weeks
• Csikszentmihalyi 1990, Flow
• Csikszentmihalyi & LeFevre 1989, Optimal experience in work and leisure
• Drucker 1967, The Effective Executive
• Ericsson, Krampe & Tesch-Römer 1993, The role of deliberate practice in the acquisition of expert performance
• Fong, Zaleski & Leach 2015, challenge–skill balance and flow (meta-analysis)
• Ghibellini & Meier 2025, Zeigarnik and Ovsiankina effects (meta-analysis)
• Harris, Allen, Vine & Wilson 2021, flow and performance (meta-analysis)
• Hartmann et al. 2021, smartphone "brain drain" (replication)
• Killingsworth & Gilbert 2010, A wandering mind is an unhappy mind
• Kushlev & Dunn 2015, Checking email less frequently reduces stress
• Leroy 2009, Why is it so hard to do my work? (attention residue)
• Mark, Gudith & Klocke 2008, The cost of interrupted work: more speed and stress
• Masicampo & Baumeister 2011, plans and unfulfilled goals
• Monsell 2003, Task switching
• Newport 2016, Deep Work
• Ophir, Nass & Wagner 2009, Cognitive control in media multitaskers
• Perlow 1999, The time famine
• Risko & Gilbert 2016, Cognitive offloading
• Rubinstein, Meyer & Evans 2001, Executive control of cognitive processes in task switching
• Ruiz Pardo & Minda 2022, smartphone "brain drain" (direct replication)
• Scullin et al. 2018, The effects of bedtime writing on difficulty falling asleep
• Storm & Stone 2014, Saving-enhanced memory
• Wiradhany & Nieuwenstein 2017, Cognitive control in media multitaskers: two replication studies and a meta-analysis

4. Energy: the machine that runs the line

In September 1915 the Minister of Munitions set up a committee to look at something nobody had much time for in wartime: fatigue. The factories were running long shifts, often seven days a week. The committee's most urgent concern was Sunday labour. Its view was that workers who never stopped did not produce more shells. They just looked more tired producing them.

A century later, an economist went back to the output records of those factories, and the committee turned out to be right. That is the theme of this chapter. The previous three stations were about what goes onto the production line, how it moves, and how much attention it gets. This one is about the machine itself, which in knowledge work is a person. Is the machine maintained, or is it being run into the ground and called dedication?

The books are full of advice here, and this is where the pattern from the introduction shows most clearly. The mechanisms mostly hold. The sizes, and a few famous findings, do not.

Sleep: the debt you don't feel

The best evidence on sleep and work comes from a small, careful lab study. Van Dongen and colleagues (2003) put 48 adults aged 21 to 38 on 4, 6 or 8 hours in bed for 14 days, and compared them with people kept awake for three nights. On 4 and 6 hours, cognitive deficits grew steadily, night after night. After two weeks, the 6-hour group was performing about as badly as people who had missed up to two full nights of sleep. The unsettling part was that their sleepiness ratings barely moved. They felt slightly tired. They were badly impaired. (~85%)

That gap between feeling and performance is the core of the problem. Short sleep is a debt that doesn't show up on your own statement. If you judge whether you're coping by how you feel, you will conclude that you are coping.

A related finding is often quoted: after 17 to 19 hours awake, people in one study performed on some tests like someone at 0.05% blood alcohol (Williamson & Feyer 2000, 39 people). It is a vivid comparison, with caveats. Long hours awake also mean late night, the body's natural low, and equal test scores are not equal crash risk.

Naps help, within limits. A meta-analysis of 11 studies found that a short nap improved cognitive performance, mostly alertness, for up to about two hours (Dutheil et al. 2021) (~85%). Almost all of this is lab work, and there's little direct evidence on whether naps can repay a chronic debt. A nap is a patch, not a repair.

This is where Matthew Walker's Why We Sleep comes in, and it deserves a fair hearing in both directions. Its core message, that chronic short sleep damages thinking and people don't notice, rests on exactly the kind of evidence above. Its headline numbers are another matter. The book says that routinely sleeping less than six or seven hours more than doubles your cancer risk; it gives no source, and a meta-analysis of 65 studies and about 1.55 million people found no link at all (Chen et al. 2018) (~5%). It says the WHO declared a sleep-loss epidemic; the WHO made no such declaration, and Walker later said he had misremembered. It says sleep time has fallen sharply over 50 years; a review of 168 studies of objectively measured sleep from 1960 to 2013 found no decline (Youngstedt et al. 2016). It even reproduces a chart with the one inconvenient bar removed. The book is right that sleep matters. It is wrong about much of the evidence it uses to make you frightened of not sleeping.

Caffeine and alcohol: the two tools that lie to you

The two most common ways people manage their energy share a property: both work on how you feel, not on what's happening.

Caffeine's half-life in a typical adult is about five to six hours, with wide variation. In one small study (Drake et al. 2013, 12 people, 400 mg), caffeine taken even six hours before bed cut objectively measured sleep by more than an hour, and the participants didn't notice (~85%). The afternoon coffee feels harmless because, subjectively, it is.

Alcohol does the reverse trick. It feels like a sleep aid and works as a sedative. A meta-analysis of 27 studies found that REM sleep is disrupted from about two standard drinks, and more with each one; falling asleep faster happened only at high doses (Gardiner et al. 2025). The glass of wine that "helps you unwind" helps you fall unconscious, which is not the same thing.

Both effects fit the Van Dongen pattern. The instrument you'd use to check your energy, your own sense of it, is the one being tampered with.

Hours: where the curve bends

Back to the munition factories. John Pencavel (2014) analysed output and hours for British munition workers, mostly women, between 1914 and 1918. Below about 49 hours a week, output rose in proportion to hours. Above that it flattened, peaking at about 63 hours. A 70-hour week produced about the same output as a 56-hour week: fourteen extra hours, nothing to show for them. Weeks without a rest day had about 10% lower output for the same hours. The Sunday committee had a point (~85% for the shape).

The obvious caveat is that turning shells on a lathe is not writing a strategy memo. The shape of the curve, rising and then flattening, probably transfers to other work. The thresholds may not (~50%).

The modern experiment runs the other way: fewer hours, not more. A study of four-day-week trials covering 141 organisations and 2,896 employees in six countries, with about five fewer hours a week, found that burnout and job satisfaction improved most, and that the gains largely held after 12 months (Fan, Schor et al. 2025). In the UK pilot of 2022, 56 of the 61 companies kept the arrangement. The weakness is in the design. The companies chose to take part, nobody was randomised, the results are self-reported, and output was barely measured; revenue stayed roughly flat. That fewer hours means less burnout is reasonably supported (~70%). That it costs no output is plausible but unproven (~50%).

Breaks and switching off

Short breaks are the most common energy advice and the most modest. A meta-analysis of micro-breaks of ten minutes or less (Albulescu et al. 2022, 22 samples, 2,335 people) found that they raised vigour and reduced fatigue by a small-to-moderate amount (d = .36 and .35). Performance barely moved (d = .16, not significant), and the gains showed up on less demanding tasks. Longer breaks helped more (~85%). So a coffee-machine break makes you feel better, which is worth having, but don't expect it to rescue a hard afternoon.

What happens after work matters more. A meta-analysis of 91 samples and over 38,000 employees found that people who mentally detach from work in their free time report less exhaustion, better sleep and higher life satisfaction, and do somewhat better on their tasks (Wendsche & Lohmann-Haislah 2017) (~70%). Two caveats. The data are mostly correlational, so healthier people may simply find it easier to switch off. And detachment went with slightly lower creativity and helpfulness, perhaps because the people who keep thinking about work are also the ones who solve problems in the shower and answer colleagues' questions in the evening.

Holidays follow the same logic and fade fast. One meta-analysis found a clear boost to wellbeing from a vacation (d = .43) that faded after return; a later one, with 13 studies, found no difference from baseline after the first week back (de Bloom et al. 2009). Longer leave seems to do more: in one study of 129 academics on sabbatical compared with 129 matched colleagues, wellbeing rose during the leave (Davidson et al. 2010).

This is the territory of Cal Newport's Slow Productivity, which argues for a "natural pace": longer timelines, and work in seasons, with intense phases followed by lighter ones. The evidence above is compatible with it, and the sabbatical study supports part of it. But nobody has tested whether seasonal rhythms beat a steady pace (~50%), and the book's star witnesses, Newton, Austen and other slow geniuses, are anecdotes picked for fit. The book's stronger argument, doing fewer things at once, is queueing math and belongs to Flow.

The willpower tank that wasn't

For about two decades, the most popular explanation of tiredness at work was ego depletion: self-control draws on a limited resource that gets used up during the day, like fuel in a tank. Resist the biscuits in the morning and you'll have less left for the report in the afternoon. A glucose drink was supposed to top the tank up. Daniel Kahneman's Thinking, Fast and Slow presented it as established; Cal Newport's Deep Work built its case for rituals on it.

It did not survive testing. A replication across 23 labs with 2,141 participants found an effect of d = .04, with a confidence interval that included zero (Hagger et al. 2016). A second, designed together with the theory's own proponents, ran in 36 labs with 3,531 participants and found d = .06; the data favoured no effect (Vohs et al. 2021). The glucose story fell with it (~5%).

The most famous field illustration went the same way. Judges were said to grant parole about 65% of the time after a food break, falling to near zero just before the next one (Danziger et al. 2011). But cases weren't heard in random order, and simulations show that sensible scheduling of cases can produce most of the pattern (Glöckner 2016).

None of this means people don't get tired. They obviously do, and the sleep and hours data above show it. What failed is the specific model: a single willpower fuel that each act of restraint drains. What replaces it is more useful. People with high self-control mostly don't resist temptation more; they run into it less. In experience-sampling studies of everyday desires, those who scored high on self-control were mainly better at avoiding tempting situations in the first place (Hofmann et al. 2012; Ent, Baumeister & Tice 2015). The practical advice in Deep Work, rituals and routines that don't need to be renegotiated every day, survives. Just not for the reason given. (How to design that environment is the subject of Choose.)

What doesn't work

• The willpower tank. Ego depletion: two multi-lab replications, over 5,600 participants combined, effects near zero (Hagger et al. 2016; Vohs et al. 2021). (~5%)
• Hungry judges as proof of depletion. The parole pattern is largely explained by how cases were ordered (Glöckner 2016). (~30%)
• Short sleep doubles cancer risk. A meta-analysis of 65 studies found no link (Chen et al. 2018). (~5%)
• Hours beyond the bend. In the munition data, 70-hour weeks produced about what 56-hour weeks did (Pencavel 2014). (~85%)
• Micro-breaks as a performance tool. They lift mood and energy, but the effect on demanding work was small and not significant (Albulescu et al. 2022). (~85%)

Try this

• Don't use how you feel to judge your sleep. Fix a bedtime that allows at least seven hours in bed for two weeks, then judge your work, not your mood.
• Move the last coffee earlier. Given a half-life of five to six hours and the Drake result, aim for no caffeine within about six hours of bed, and more if you're sensitive.
• Treat evening alcohol as a sleep cost. From about two drinks, expect worse REM sleep, whatever it feels like.
• Cap the week and keep a full day off. If you're regularly past 50 hours, the extra hours are likely buying very little; a weekly rest day was worth about 10% of output in the munition data.
• End work deliberately. Close the laptop at a set time and stop checking messages, so that the evening is actually off. Use short breaks for energy, not as a fix for overload.

Sources

• Albulescu et al. 2022, micro-breaks meta-analysis (PLOS ONE)
• Chen et al. 2018, sleep duration and cancer meta-analysis (BMC Cancer)
• Danziger et al. 2011, parole decisions and food breaks
• Davidson et al. 2010, sabbatical leave and wellbeing
• de Bloom et al. 2009, vacation effects meta-analysis
• Drake et al. 2013, caffeine taken before bedtime
• Dutheil et al. 2021, short naps and cognitive performance meta-analysis
• Ent, Baumeister & Tice 2015, self-control and avoiding temptation
• Fan, Schor et al. 2025, four-day week trials (Nature Human Behaviour)
• Gardiner et al. 2025, alcohol and sleep meta-analysis (Sleep Medicine Reviews)
• Glöckner 2016, the hungry-judge effect revisited
• Hagger et al. 2016, multi-lab replication of ego depletion
• Hofmann et al. 2012, everyday desire and self-control
• Pencavel 2014, the productivity of working hours (IZA DP 8129)
• Van Dongen et al. 2003, chronic sleep restriction (Sleep)
• Vohs et al. 2021, multi-site test of ego depletion
• Wendsche & Lohmann-Haislah 2017, detachment from work meta-analysis (Frontiers in Psychology)
• Williamson & Feyer 2000, sleep deprivation vs. alcohol intoxication
• Youngstedt et al. 2016, trends in adult sleep duration

5. Leverage: what multiplies an hour

Archimedes is supposed to have said that with a place to stand and a long enough lever he could move the earth. The line was written down centuries after his death, so nobody knows whether he said it. That makes it a fitting start for this chapter: like much productivity advice, it pairs a real mechanism with a size nobody checked.

The first four stations are about not wasting input: choosing the right work, keeping it moving, giving it attention, keeping the machine running. This station is about the other half of the ratio. Leverage is anything that makes an hour of input produce more valuable output: a skill built over years, a tool, an AI model, a colleague you trained. Every lever costs something to build or use, and some of them quietly push the wrong way.

The question for this station: what actually multiplies output, and by how much?

AI: a jagged frontier

Ethan Mollick's Co-Intelligence describes AI as a strange new co-worker with a "jagged frontier": brilliant at some tasks, oddly bad at others that look just as hard, and you can't tell which is which in advance. The best evidence for this comes from a study Mollick co-authored. 758 consultants at Boston Consulting Group worked on 18 realistic tasks. With GPT-4 they finished 12.2% more tasks, 25.1% faster, at over 40% higher quality (Dell'Acqua et al. 2026) (~85%). But on one task built to sit just outside the model's ability, 84.5% of consultants without AI got the right answer, against 60.0% and 70.6% in the two AI groups (~70%). The only difference was which side of an invisible line the task fell on.

Other studies point the same way inside the frontier. In a randomised trial with 453 professionals doing short writing tasks, ChatGPT cut the time by 40% and raised quality by 18% (Noy & Zhang 2023). Among 5,172 customer-support agents, an AI assistant raised issues resolved per hour by 15% (Brynjolfsson, Li & Raymond 2025). In the working paper, novice and low-skilled agents gained 34%.

That last detail is the most consistent finding in the field: AI levels skill (~85%). In the BCG study, below-average consultants improved quality by about 43%, top performers by about 17%. Experienced support agents gained little speed and lost slightly on quality. A plausible mechanism: the model supplies a competent average answer, which is a big step up for a beginner and a small one, or a step sideways, for an expert.

The catch is that a good assistant makes people stop checking. In a preregistered experiment, 181 recruiters scored résumés with an AI that was about 99%, 85% or 75% accurate, or with none. Those given the better AI put in less effort, followed it more blindly and judged less accurately (Dell'Acqua 2022). Mollick calls this falling asleep at the wheel. The better the assistant, the more tempting it is to stop reading its work.

Code: the stopwatch and the feeling disagree

In the most-quoted trial of AI for programmers, 95 freelance developers recruited on Upwork wrote a small web server in JavaScript. Those with GitHub Copilot finished in 71 minutes instead of 161, 55.8% faster (Peng et al. 2023). The result is solid for that task (~85%) and a weak guide to real work (~30%): one bounded job, written from scratch, with the less experienced gaining most. Field trials at Microsoft, Accenture and a Fortune 100 firm, covering 4,867 developers, found about 26% more completed tasks, again mostly for juniors (Cui et al. 2025).

Then METR looked at a different population: 16 experienced open-source developers working on 246 real tasks in codebases they knew well. With early-2025 AI tools they took 19% longer. Before starting they had forecast a 24% speed-up, and afterwards they still believed they had been about 20% faster (Becker et al. 2025) (~85%). The slowdown belongs to one setting and one generation of tools. METR's 2026 follow-up was abandoned as unreliable, partly because developers refused to work without AI, and secondary reports put the new estimates near zero. The gap between feeling and stopwatch is the part worth keeping, and Feedback returns to it.

Teams show a similar puzzle. Google's DORA survey for 2024 found that 25% more AI adoption went with about 1.5% lower delivery throughput and 7.2% lower stability, even as code quality and review speed improved slightly (~70%). The 2025 report found AI now linked to higher throughput but still to more instability, with code review as the new bottleneck. Both are survey associations, not experiments. The mechanism is familiar from the factory floor: speeding up one step only moves the pile to the next one (see Flow).

None of this would have surprised Fred Brooks. In "No Silver Bullet" (1986) he argued that no single technology would deliver even a tenfold gain in software productivity within a decade, because most of the difficulty is essential (the complexity of the problem itself) rather than accidental (friction from tools and languages) (~85%). AI tools are the toughest test his claim has faced, and none of the controlled studies comes close to tenfold. To be fair to the enthusiasts, Mollick's advice to assume "this is the worst AI you will ever use" also has data behind it: the length of task AI agents can complete doubled roughly every seven months from 2019 to 2025. Brooks is still right, but the margin is shrinking.

Skill: the slow lever

Skill is the oldest lever, and the one most distorted by its popularisers. Anders Ericsson's research on violinists at a Berlin music academy found that the best students had practised alone far more than the merely good ones: about 7,410 hours by age 18, against 5,301 (Ericsson, Krampe & Tesch-Römer 1993). Malcolm Gladwell turned the study's average for the best group at age 20 into the "10,000-hour rule". In Peak, Ericsson rejects the rule himself: there is no magic number, and the hours differ by field.

The bigger problem is with Ericsson's own claim. A double-blind replication with 39 violinists found that practice explained about 26% of the differences in skill, not 48%, and the "good" group had practised more than the "best", though not significantly (Macnamara & Maitra 2019). A meta-analysis of 88 studies found that deliberate practice explained 26% of performance differences in games, 21% in music, 18% in sports, 4% in education and under 1% in professions (Macnamara, Hambrick & Oswald 2014) (~85%). Practice matters. It is not the whole story, and in most jobs it is a small part of it.

What survives is the mechanism. Practice builds mental representations: chess masters recall real game positions at a glance but lose most of that advantage when the pieces are placed at random (Chase & Simon 1973). And experience alone is not practice: without feedback, years on the job can even make people worse (see Feedback).

Knowledge workers rarely have what deliberate practice requires: an established training method and a coach who designs exercises. So Cal Newport's suggestion in So Good They Can't Ignore You that deliberate practice will let most knowledge workers leap ahead rests on thin evidence. The famous "10× programmer" has the same problem: the 1968 study behind it mixed languages and methods, and a re-analysis attributes only about 5:1 to the programmers themselves. Large differences between people are real. Where they come from is less clear than the books suggest.

AI now sits right on this lever. Adam Grant's Hidden Potential argues for scaffolding: temporary support, like a tutor or worked example, that gets a learner past an obstacle and then fades. A randomised trial with about 1,000 Turkish high-school students shows what happens when it doesn't fade. Plain GPT-4 access raised practice scores but cut exam scores by 17%; a version built as a tutor, with guardrails, did no harm (Bastani et al. 2025) (~85%). Used as an answer machine, AI borrows today's output against tomorrow's skill.

Other people: the manager's lever

Andy Grove, Intel's long-time chief executive, built High Output Management on the same ratio as this essay. A manager's output, he wrote, is the output of their organisation plus the neighbouring groups they influence, and leverage is an activity's effect on that output divided by the time it takes. High-leverage work affects many people, lasts a long time, or uses knowledge only you have.

His favourite example is training: a few hours of preparation pay back across a team's whole year. The evidence that training works is good. A meta-analysis found an effect of about d = 0.62 on on-the-job behaviour (Arthur et al. 2003). That it beats a manager's other uses of time is Grove's arithmetic, not a finding. In a randomised trial in Indian textile plants, introducing basic management practices raised productivity by about 17% in the first year (Bloom et al. 2013). That was a factory, though, and moving the result to an office is an extrapolation.

Grove's people advice ages less well. His "task-relevant maturity", which matches management style to a person's experience with the task, is a version of situational leadership theory, which has fared poorly in tests. His warning that "delegation without follow-up is abdication" is untested but sensible. It also reads well as a guide to delegating to AI agents: a delegate who has never done the task needs checking, whoever or whatever it is. Adding people has its own coordination cost, which belongs to Flow.

Where the work happens

Remote work is a lever many people now argue about and few have measured. The best evidence is two randomised trials at the same Chinese travel company. In the first, call-centre volunteers allowed to work from home for nine months performed 13% better: 9% from working more minutes, 4% from more calls per minute. Quitting halved, though promotion rates fell for the same performance (Bloom et al. 2015). The result is solid for routine, individual work (~85%) and much less certain beyond it (~50%). One co-author co-founded the company.

The second trial, with 1,612 engineers, marketing and finance staff, tested two days a week at home over six months. Quitting fell by a third, with no effect on performance reviews or promotions (Bloom, Han & Liang 2024) (~85%). Managers had predicted a productivity loss beforehand. The honest summary: where the work happens is a weak output lever and a strong retention lever. The debate is louder than the effect.

What doesn't work

• The 10,000-hour rule. No threshold exists; the number was one group's average at age 20, and Ericsson called it arbitrary (~95% that the rule is a misreading).
• "Practice, not talent." In a study of about 10,500 Swedish twins, identical twins who practised different amounts did not differ in musical ability (Mosing et al. 2014) (~15% for the book's claim).
• Persona prompts. Telling an AI "you are an expert" did not raise accuracy across 162 roles and 2,410 questions; context and examples do help (Zheng et al. 2024) (~30% for personas, ~70% for context).
• Lab speed-ups as real-world gains. The 55.8% Copilot result came from one bounded task; experienced developers in their own codebases were slower (~30% that the lab figure carries over).

Try this

• Map your own frontier with a stopwatch. Pick a recurring task, do it a few times with AI and a few without, and time both. In the METR study, felt and measured speed were about 40 percentage points apart.
• Check hardest where AI helps most. It helps beginners most, and beginners are worst at spotting its errors. In unfamiliar areas, verify facts and sources before using the output.
• When learning, make AI the tutor, not the answer key. Ask it for hints, questions and explanations; do the final step yourself.
• Practise the part of your work that has feedback. Pick one skill where you can see results quickly, work at its edge for short, focused sessions, and find someone who can design exercises for it.
• Spend an hour on something that lasts. A checklist, a training session or a written answer to a recurring question pays back every time someone uses it.

Sources

• Arthur, Bennett, Edens & Bell, 2003, meta-analysis of organisational training
• Bastani et al., 2025, GPT-4 access and learning in Turkish high schools (PNAS)
• Becker et al. (METR), 2025, AI and experienced open-source developers
• Bloom et al., 2013, management practices in Indian textile plants (QJE)
• Bloom, Liang, Roberts & Ying, 2015, working from home at Ctrip (QJE)
• Bloom, Han & Liang, 2024, hybrid work at Trip.com (Nature)
• Brooks, 1986, "No Silver Bullet" (in The Mythical Man-Month, 1995 ed.)
• Brynjolfsson, Li & Raymond, 2025, AI and customer-support agents (QJE)
• Chase & Simon, 1973, perception in chess
• Cui et al., 2025, field experiments with AI coding assistants (Management Science)
• Dell'Acqua, 2022, recruiters and AI accuracy
• Dell'Acqua, McFowland, Mollick et al., 2026, the jagged frontier at BCG (Organization Science)
• DORA, 2024, State of DevOps report
• Ericsson, Krampe & Tesch-Römer, 1993, deliberate practice in violinists
• Ericsson & Pool, 2016, Peak
• Grant, 2023, Hidden Potential
• Grove, 1983, High Output Management
• Macnamara, Hambrick & Oswald, 2014, deliberate practice meta-analysis (Psychological Science)
• Macnamara & Maitra, 2019, violinist replication
• Mollick, 2024, Co-Intelligence
• Mosing et al., 2014, music practice in Swedish twins
• Newport, 2012, So Good They Can't Ignore You
• Noy & Zhang, 2023, ChatGPT and professional writing (Science)
• Peng et al., 2023, GitHub Copilot experiment
• Zheng et al., 2024, persona prompts in LLMs

6. Feedback: how to read the gauges without fooling yourself

Daniel Kahneman once taught flight instructors in the Israeli Air Force that rewarding good performance works better than punishing mistakes. A senior instructor objected. When he praised a cadet for a clean manoeuvre, he said, the next attempt was usually worse. When he shouted at a cadet for a sloppy one, the next attempt was usually better. Praise didn't work; shouting did. He had years of experience to prove it.

He was describing something real and explaining it wrongly. A cadet who flies an unusually good manoeuvre has probably had some luck, and the next attempt will usually be closer to their average, praise or no praise. The same goes in reverse for an unusually bad one. Kahneman had the instructors throw coins at a target twice. Those who did best the first time mostly did worse the second, and nobody had shouted at anyone.

This chapter is about the last station on the line, the one that tells you whether the other five are working. It sounds like the easy one: count something, look at the number, adjust. The flight instructor shows why it isn't. Feedback can make you better, and it can just as easily make you confidently wrong.

Count results, not motion

Most of what is easy to count is activity: hours logged, emails sent, tickets closed, lines of code written. Accelerate (Forsgren, Humble & Kim 2018) makes the case that these are the wrong things to watch. They measure motion rather than results, they are easy to game, and pushing utilisation towards 100% makes queues explode (~85%; the queueing maths is in Flow).

The book's alternative, built from the State of DevOps surveys, is four measures of software delivery: how often a team deploys, how long a change takes to reach users, how quickly service is restored after a failure, and how often changes fail. These "DORA metrics" became the industry standard, and they hang together as a coherent picture of delivery performance. The most interesting finding is that speed and stability didn't trade off. The best teams were faster and broke things less (~70%). The caveat: these are self-reported surveys from self-selected enthusiasts, so they show that good teams have both, not that speed causes stability.

The authors later went further. The SPACE framework (Forsgren, Storey et al. 2021) argues that no single number captures developer productivity, and recommends at least three of five dimensions (satisfaction, performance, activity, communication, and efficiency or flow), with at least one that asks people how things feel. It is a framework rather than a tested model, but its core point travels well beyond software.

For a single person the translation is simple: ask what the week produced, not how busy it was. A thesis chapter drafted, a patient discharged safely, a customer who renewed: those are output. Everything else is input in disguise.

Every number needs a rival

In 2016 it emerged that employees at Wells Fargo, under pressure to hit targets for new accounts per customer, had opened millions of accounts that customers never asked for. The bank fired about 5,300 people and paid a $185 million fine. The account count went up. The thing it was supposed to signal, customers who wanted more from the bank, went down.

This is Goodhart's law, in the anthropologist Marilyn Strathern's popular phrasing: when a measure becomes a target, it ceases to be a good measure. Economists have a formal version. Holmström and Milgrom (1991) showed that when you reward only the part of a job that is easy to measure, effort flows away from the parts that are hard to measure. Nobody needs to be a villain for this to happen; they only need to respond to what is counted.

Andy Grove's fix in High Output Management is to pair every indicator with its counter-indicator: measure an effect and the thing that would suffer if someone over-optimised it (~85%). Count output, and also count defects. Track speed, and also track rework. The DORA metrics are built this way, two for speed and two for stability, which is why gaming one tends to show up in another. Grove also prefers leading indicators, which warn of trouble before it reaches the output; the hard part, which no study solves, is finding one that really leads.

For a person, a pair might be "pages written" and "pages kept after a week", or "tasks closed" and "tasks reopened". One number can be flattered. Two numbers pulling in opposite directions are much harder to fool.

Noise in the signal

Back to the flight instructor. Regression to the mean is not a psychological quirk but arithmetic: whenever performance has a random component, an extreme result tends to be followed by a less extreme one (~95%). The trap is that the follow-up result arrives right after whatever you did in response, so you credit or blame your response. Shout after a bad week and the next week usually improves. Celebrate a great week and the next one disappoints.

Small samples make it worse. Kahneman's example: the counties with the lowest kidney-cancer rates are mostly small and rural, and so are the counties with the highest. Small counties are simply too small to produce stable averages, so they land at both extremes. A single person's week is a very small county. One brilliant Tuesday or one wasted Thursday says little, yet it's tempting to build a new system on either.

The same noise is why decisions should be judged by their process rather than their outcome. Under uncertainty a good decision can turn out badly and a reckless one can get lucky, yet people judge by results anyway. Baron and Hershey (1988) showed this outcome bias, and a preregistered replication with 692 participants confirmed it, even among people who said outcomes shouldn't count (Aiyer et al. 2023). Whether reviewing your process actually improves later decisions is much less tested. The tools for making decisions in advance live in Choose; the point here is simply that a result is a noisy grade on a decision, not a verdict.

Loops that actually teach

Peak (Ericsson & Pool 2016) is a book about deliberate practice, which belongs to Leverage. But it makes one point that belongs here: without feedback, practice doesn't improve anything. Purposeful practice, in the book's terms, needs a specific goal, full focus, immediate feedback and work just beyond what is comfortable. Remove the feedback and repetition turns into rehearsal of whatever you already do.

The most sobering evidence comes from medicine. A systematic review of 62 studies found that over half reported doctors' performance declining with years in practice, and only one found improvement (Choudhry, Fletcher & Soumerai 2005) (~70%). The caveats are real: most studies measured knowledge and adherence to guidelines rather than patient outcomes, and older doctors trained under older guidelines. Still, the pattern fits a broader rule. Kahneman and Klein (2009) argue that expert intuition can be trusted only where the environment is regular and feedback is fast and clear, as in chess or firefighting, and not where it is slow or murky (~70%). Twenty years of experience can be one year repeated twenty times if nothing ever tells you which of your calls were wrong.

Feedback can also make things worse. A large meta-analysis by Kluger and DeNisi (1996), covering 607 effect sizes, found that feedback improves performance on average, but over a third of feedback interventions decreased it, especially when the feedback drew attention to the person rather than the task. "Your report missed the deadline because the data arrived late" helps. "You're not a details person" mostly doesn't.

The simplest loop of all, watching your own progress, holds up well. A meta-analysis of 138 randomised studies with nearly 20,000 participants found that monitoring progress improved goal attainment, with larger effects when progress was physically recorded or made public (Harkin et al. 2016) (~85%). Most of the goals were about health, but the mechanism is general. This is the part of Atomic Habits' advice to track your habits that rests on solid ground.

The review

David Allen calls the weekly review the critical success factor of Getting Things Done: without it, the lists go stale and the system rots. It's a plausible claim, and almost untested (~50%). The closest evidence is about reflection on learning. In a study of call-centre trainees at Wipro in India, those who spent 15 minutes at the end of each day writing about what they had learned scored about 23% higher on the final test (Di Stefano et al. 2014). That supports the habit of looking back more than Allen's specific ritual.

A review has a hidden enemy: memory. Kahneman's research on the "remembering self" found that people judge a past experience mostly by its most intense moment and its end, and barely by its length (~85% for remembered pain, ~70% as a general rule). In a trial with 682 colonoscopy patients, a gentler ending made the whole procedure remembered as less unpleasant. A week reviewed from memory is a week reviewed by its worst meeting and its Friday afternoon. Written notes from the week itself are the antidote. So is the reverse use: how a work session ends shapes whether you want to return to it, which is a decent argument for stopping on something that went well.

The 37× problem

Atomic Habits (Clear 2018) offers the most quoted number in the genre: get 1% better every day and after a year you'll be about 37 times better. The arithmetic is correct: 1.01 to the power of 365 is about 37.8, and 0.99 to the power of 365 is about 0.03 (~95%). As a description of how people actually improve, it is wrong (~15%).

Real learning curves have the opposite shape. The power law of practice (Newell & Rosenbloom 1981) describes large gains early and shrinking returns later. Your first week of learning to type, cook or write code brings more improvement than your hundredth. Clear's related idea, the "plateau of latent potential", in which results stay invisible for a long time and then break through, has the curve backwards for most skills (~30%). Some outcomes have thresholds, but there is no general law of hidden progress.

The book gets one thing right: small, repeated improvements add up. They add up the way practice does, quickly at first and then slowly, not the way a savings account does. The practical danger of the 37× story is in how you read your own feedback. If you expect steady compounding, a slowdown in month three looks like failure. It's usually just the normal shape of learning.

What doesn't work

• "1% better every day = 37× in a year." The maths holds; real progress follows a curve of diminishing returns (Newell & Rosenbloom 1981). (~15% as a description of progress)
• The "plateau of latent potential." Skill gains come early, not after a long invisible wait. (~30%)
• Judging a decision by how it turned out. Outcomes are noisy; people still do it, even when they say they shouldn't (Aiyer et al. 2023). (~85% that the bias is real)
• Feeling productive as a measure. Developers using early-2025 AI tools felt about 20% faster while taking 19% longer (METR 2025; see Leverage). (~85% for the perception gap)
• Single activity metrics. Lines of code, hours and utilisation get gamed and pull effort away from what isn't counted (Holmström & Milgrom 1991). (~85%)

Try this

• Write down what the week produced, not what it consumed. Two or three results, in a sentence each, every Friday.
• Pair every number you track with its rival: output with rework, speed with errors, pages written with pages kept.
• Don't change your system after one great or terrible week. Look at four to eight weeks before deciding something works or doesn't.
• Record progress where you can see it, on paper or a shared board, rather than trusting your memory of the week.
• When you review a decision, grade the process separately from the result: what did you know, what did you weigh, what would you do again with the same information?

Sources

• Aiyer et al., 2023, outcome bias replication
• Allen, 2001, Getting Things Done
• Baron & Hershey, 1988, Outcome bias in decision evaluation
• Becker et al. (METR), 2025, AI and experienced open-source developers
• Choudhry, Fletcher & Soumerai, 2005, Experience and quality of care (systematic review)
• Clear, 2018, Atomic Habits
• Di Stefano, Gino, Pisano & Staats, 2014, reflection and learning (HBS working paper 14-093)
• Ericsson & Pool, 2016, Peak
• Forsgren, Humble & Kim, 2018, Accelerate
• Forsgren, Storey et al., 2021, The SPACE of developer productivity
• Grove, 1983, High Output Management
• Harkin et al., 2016, Monitoring goal progress (meta-analysis)
• Holmström & Milgrom, 1991, Multitask principal-agent analyses
• Kahneman, 2011, Thinking, Fast and Slow
• Kahneman & Klein, 2009, Conditions for intuitive expertise
• Kluger & DeNisi, 1996, Feedback interventions (meta-analysis)
• Newell & Rosenbloom, 1981, Power law of practice

7. What's left

Strip 23 productivity books down to the claims that survive, and you are left with something smaller than any of them promised and sturdier than most of them delivered.

It isn't a system. Nobody who reads the evidence will come away believing in one weird trick, a morning routine or a 37-fold year. What remains is closer to an engineer's checklist: six stations on a production line, each with a few rules that rest on mathematics or on findings that have been replicated, and a longer list of things that sound right and aren't.

The line on one page

• Choose: Decide in advance: specific goals (learning goals for new work), simple rules for recurring decisions, and an explicit list of what you'll neglect. Best evidence: Goals vs. "do your best", d = .42–.80 (~85%); rules beat expert judgment in 136 studies (~95%)
• Flow: Limit work in progress, leave slack, find the bottleneck, and estimate from history, not hope. Best evidence: Little's law (~95%); waits explode near full utilisation (~85%); the planning fallacy (~95%)
• Attention: Protect blocks for work that needs them, by rule rather than willpower; offload everything else into a system you trust. Best evidence: Switching costs (~70%); offloading frees memory (~85%); quiet time works while enforced (~85%)
• Energy: Sleep enough before judging anything else; cap hours; actually stop. Best evidence: Chronic 6-hour sleep impairs without feeling it (~85%); output flattens with long hours (~85%)
• Leverage: Use tools and AI inside their frontier, check them outside it, and build skill where feedback exists. Best evidence: AI gains inside the frontier, losses outside (~85%); practice explains a minority of performance (~85%)
• Feedback: Count results, not motion; pair every metric with its rival; judge over weeks, not days. Best evidence: Goodhart and paired indicators (~85%); regression to the mean (~95%); monitoring progress (~85%)

Read the right-hand column again. The strongest entries are an identity from queueing theory, an estimation bias replicated for decades, a statistical inevitability and a meta-analysis about formulas. Not one of them is inspiring. All of them work whether or not you believe in them, which is more than can be said for the willpower tank.

Three lessons from the checking

Halve the effect, keep the mechanism. The single most common verdict was "right direction, wrong size". If-then plans help, at perhaps a quarter to half the size the books quote. Temptation bundling helps, at a fraction of its first experiment. Rewards dampen interest, but don't kill creativity. A useful reading habit follows: when a productivity book reports an effect, assume the mechanism is probably real and the number is probably the best one the author could find.

Prefer arithmetic to inspiration. The advice that moves people is mostly psychology from before the replication crisis, and that is exactly where the failures cluster: willpower that runs out, phones that drain the brain by sitting there, multitasking that rewires the mind, practice that explains everything. The advice that holds is mostly arithmetic: queues, bottlenecks, coordination paths, regression. It's dull to read and reliable to apply, and two of the three highest-rated books in the collection are novels about factories and IT departments.

Don't trust the gauge on the dashboard. This one runs through every chapter without belonging to any of them. People sleeping six hours a night felt slightly tired while performing like people who had missed whole nights. A coffee six hours before bed cost an hour of sleep nobody noticed. A glass of wine that feels like a sleep aid quietly disrupts the sleep it seems to bring. Experienced developers believed AI had made them 20% faster while the stopwatch said 19% slower. And a week remembered is a week judged by its worst moment and its ending. The instrument most people use to steer their productivity, their own sense of how it's going, is the least reliable one on the line. Most of the surviving advice is, in the end, a way to replace that gauge with something better: a bedtime, a calendar, a WIP limit, a stopwatch, a written note from the week itself.

The hall of shame

Some ideas are worth knowing mainly so you can stop believing them:

• The willpower tank. Two multi-lab replications, more than 5,600 people, effects near zero (~5%).
• "Forget goals, build systems." Goal research says the opposite (~30%).
• "1% better every day = 37× in a year." True arithmetic, false description of learning (~15%).
• The 10,000-hour rule. An average from one study of violinists, rejected by its own source.
• "It takes 23 minutes to refocus." Not in the paper usually cited, which found people working faster (~30%).
• The phone that drains your brain by lying there (~15%) and multitasking that ruins focus (~15%).
• "Rewards kill creativity." The showcase experiment didn't replicate (~15%).
• Short sleep doubles cancer risk. No link in 65 studies (~5%), and the book's case for sleep never needed it.

None of these were fringe claims. Each appeared in a bestseller, several in more than one, and they are the claims readers remember best. A reasonable test for any new piece of advice is whether it would make this list in five years.

Where to start

The line is only as fast as its slowest station, which is the one lesson the books agree on most. So don't adopt all six chapters. Find the station where your own line is worst. If you're exhausted, it's Energy; nothing downstream works on a short night. If everything is "in progress" and nothing finishes, it's Flow. If you're busy but unsure the work matters, it's Choose. Fix that one, watch the results for a month rather than a week, then look for the next bottleneck. That is Goldratt's method, applied to a person.

A minimal version of the whole line fits on an index card:

1. Write down what you're aiming for, and what you're not doing.
2. Have no more than three things in progress.
3. Book the blocks that matter; let the rest of the day absorb the surprises.
4. Protect seven hours in bed, and stop work at a set time.
5. Time it before you believe a tool makes you faster.
6. Every Friday, write what the week produced, not what it consumed.

A version number

Kahneman's priming chapter is a reminder that careful people get things wrong and that the honourable response is to say so. This essay will be wrong in places too. The ratings are judgements, the shelf was one person's, and several findings here, the AI studies above all, will look different in a year.

So treat this as version 1. The claims are rated so that they can be re-rated. When the evidence moves, the verdicts should move with it, and the line should be rebuilt from whatever is still standing. That is, after all, what the Feedback station is for.