A PM at a Series-B startup ships a feature on Tuesday. He describes the change in plain English to Claude Code. The model writes the code, runs the tests, fixes what failed, and opens the pull request. He has never written a line of code before.
A founder builds a working MVP of her next product over a weekend in Cursor, alone at her kitchen table. Two years ago that prototype would have cost a hundred thousand dollars, taken half a year, and required a team of four. Today it is one person and one weekend.
Agents now perform many of the Jobs that used to need whole teams. Writing the code, drafting the strategy memo, building the marketing deck, running customer outreach — each of those once meant a payroll line and a hiring cycle, and each now is possible during a chat session.
This is subtraction running faster than any working generation has seen it run. For a huge range of customer Jobs, agents can now climb a level above the old way of performing them, and the chain of work underneath shrinks dramatically. The operation that built the modern world is now running fast enough to watch in real time.
The Big Jobs Stay While the Graph Underneath Collapses
Take any Big Job a person performs in 2026 and rewind it a hundred years. The Big Job has not changed, but the Job Graph underneath it has collapsed beyond recognition.
Start with feed my family this week. That Big Job has been the same for dozens of thousands years.
In 1926 it meant walking to the market most days, because nothing kept. Meat came home wrapped in paper, and the dried beans soaked overnight. The iron stove needed a fire built in it before anything could cook, and yesterday's scraps got rendered into fat because nothing was thrown away. Between the chopping, salting, boiling, watching, and turning, the Graph under that one weekly Job ran hundreds of Jobs deep.
Then Solutions started deleting Jobs from it, decade by decade. By 1955 the refrigerator had killed go to the market every day, and frozen food ended the nightly puzzle of plan a meal around what spoils first. By 1985 the microwave erased thaw on the stovetop for an hour: a frozen cut that once sat on the counter for four hours, or took forty minutes of attended turning in a low pan, came back in ten minutes. Cooking shifted from every evening to once a week. By 2005 supermarket pre-cut produce had taken over the trimming and chopping, and by 2015 meal kits absorbed the menu, the list, and the trip. By 2018 a tap on a phone climbed above cooking entirely — a hot meal at the door in twenty minutes.
The same compression runs through every other Big Job in a person's life. Get from where I live to where I need to be once meant owning a horse, tolerating weather, and planning around daylight. Now it means opening an app and watching a car arrive in four minutes.
The Jobs at the very top of a person's life barely change at all. Feed my family, own a home, get around — people have used those verbs for centuries. The Jobs underneath them disappear by the decade: untangle the headphones, type the password, carry a folded paper map, calculate the route from printed directions. All gone.
Two Mechanics Do Most of the Work
The full catalog holds about a hundred ways to create customer value and business value. Two of them beat all the others by a wide margin, and both work by taking things away.
The first is Kill a Job: a Job disappears from the customer's day as a class. AirPods killed untangle my headphones, Face ID killed type my password, Apple Pay killed reach for my wallet, and automatic bill-pay killed remember to send the credit-card payment by the 12th. Each of those was a Job the customer performed dozens of times a day. Multiply by a billion users and the count of removed actions gets hard to write down.
Kill-a-Job is the atomic move — the single verb under most of the rest of the catalog. Fix the Problem kills the Job the Problem keeps regenerating. Take the Job off the customer is the same verb aimed at one customer at a time. Reduce the wait deletes the Job of waiting for the previous step to finish, and remove a negative emotion deletes the Job that produced it. Once the verb is visible, much of the value-creation catalog reads as one operation under different names. So does a large share of strategy: subtract Jobs from the target segment's Graph, subtract entire non-target segments from the team's investment landscape, subtract C and D customers, subtract risky assumptions.
The second is Move up a level: take the Big Job one rung above your current Core Job and make it your new Core Job. This is Kill-a-Job at scale. The moment you deliver the Big Job turnkey, every lower-level Job the segment used to perform on the way to it collapses out of the Graph.
Uber climbed above own a car, and the rider stopped buying, financing, insuring, refueling, parking, registering, and selling a vehicle — an entire branch of Jobs killed by a single climb. Squarespace did the same to configure cPanel, install WordPress, renew the SSL; TurboTax to read the IRS instructions, pick the right forms, check the arithmetic; ChatGPT to find a template, format the headings, draft from a blank page, restructure later.
Both mechanics shrink the customer's life, the first one Job at a time, the second a whole branch at once. Every other move in the catalog works on the same underlying shape: it finds something the customer was paying with and hands it back.
One Operation Under the Whole Catalog
On a slide, the mechanics look like a list of different things. Walked through one by one, they turn out to be the same thing seen from different angles. They are all removals.
Step back one more level and the pattern widens. The four pillars of the larger methodology — AJTBD, Unit Economics, RAT, ABCDX — are four shapes of one operator. AJTBD removes Jobs from the customer's day, and entire non-target segments from the team's investment landscape. Unit Economics strips out negative-margin customers, channels, and products. RAT clears away risky assumptions before they get built into a product the market would refuse, and ABCDX cuts the C and D customers whose cost-to-serve eats the margin earned on the A's and B's. One operation, four shapes, and its name is subtraction.
A full treatment of subtraction as the meta-operator across all four pillars belongs to the next book. What matters here is that every chapter of this one has been pointing at it without naming it.
I made a claim that sounded soft at the time and carries the ending now. The customer does not need your product; the customer needs a Job performed. Take that seriously enough and the most valuable product becomes the one that is never present at all — the Job performed, finished, gone, and the customer standing at the State B they wanted with nothing in their hand.
Apple: Thirty Years of Subtraction
Two companies in the current generation run this operation at industrial scale, and Apple has the longer record.
In 2007 Apple released a single device that absorbed three. The iPod carried your music, the phone made your calls, and the browser was how you read the web anywhere other than a desk. All three collapsed into one rectangle in the pocket: three artifacts gone from the customer's day, one in their place. That climb above carry the devices I need today is arguably the most lucrative product move in modern industrial history. It was a subtraction before it was an addition.
The trail continued for another fifteen years. Face ID killed type my password and Apple Pay killed reach for the card. Universal Clipboard killed email myself the text from my Mac to my iPhone, and with it the boundary between two devices. AirPods killed untangle the headphones, then took the wire away entirely. Each move removed a small Job the customer performed a dozen times a day. After thirty years of this, Apple's product line asks for the smallest daily set of actions in consumer electronics.
Tesla Runs the Operation at Three Altitudes
The other company runs the same operation on a different surface, and the cleanest way to see it is at three altitudes inside one company.
The first altitude is the Supercharger. A Tesla pulls into a stall on a Tuesday afternoon. The driver gets out, lifts the cable off the post, and brings the handle toward the car; the charge port slides open on its own. The driver pushes the handle home, hears the soft click, and walks off toward a coffee shop. The phone never left the pocket — no app opened, no card swiped, no PIN typed, no screen asking for a payment method. The car already knew which car it was, the network already knew which account to bill, and charging started in the time it takes to read this sentence.
At another brand's fast-charger ten miles away, the same Tuesday afternoon, the driver is holding the phone. There is an app to open, a session that times out, a password to retype, a stall number to tap, and a payment method to confirm. The cable goes in, the app comes back out, the "Start charging" button hides somewhere new, and "Authorizing" hangs on the screen until the driver retries it. Same Job, same outcome, roughly the same price. What differs is how much of the customer's day each network makes the driver walk through.
The second altitude is the driving itself. With Full Self-Driving engaged, the driver taps a destination on the screen and the car takes the trip from there: it backs out of the driveway, makes the residential left, waits at the four-way, merges onto the freeway, takes the exit, threads the city traffic, waits at the red, and finds a spot in the lot and parks. The driver sits with hands in lap, and the foot never touches a pedal. Steer, brake, accelerate, change lanes, watch the gap, take the turn, find a spot, park — eight Jobs every driver has performed dozens of times an hour for a century, gone in one move. The customer is still in the car and is no longer driving it.
The third altitude is the car itself. A Waymo pulls up at a curb in San Francisco with no driver in it. The passenger gets in the back, rides across town, and gets out at the destination — having never bought the car, financed it, insured it, parked it, charged it, or touched the wheel.
The Tesla Robotaxi service in Austin runs the same operation. The Big Job — get from where I live to where I need to be — has carried the same verb for a hundred years, and today it performs through a chain small enough to fit in one clause: open the app, wait four minutes. The removals Tesla was making one Job at a time inside car ownership get made all at once by climbing to the level above ownership. Uber pointed at this fifteen years ago by climbing above own a car; Robotaxi finishes the climb by removing the driver too. Three altitudes, one company, one operation.
The Bias Against Subtraction Is Wired In
And yet teams almost never run the operation. A research team studying how people improve things ran eight experiments across LEGO structures, miniature golf courses, written essays, and travel itineraries. Asked to improve each one, sixty to eighty percent of participants reached for an addition, and the smaller fraction that reached for a removal produced the stronger outcomes. The shape held across four very different artifacts. The reflex to add is wired in, not learned, and it produces weaker work than the removals it crowds out.
The institutional version of the same bias compounds it. Addition shows up in the demo, the changelog, the press release, and the performance review. Subtraction shows up nowhere, so the team that ships fewer things looks like the team that did less work. Multiply that across quarters and the portfolio fills with additions while the removals that would have produced more value never get scheduled.
There is also a hierarchy to it. The local optimum is delegable: a junior PM can add a Job below, fix a Problem, lift a conversion rate, add a channel. The global optimum is subtractive — change the segment, kill the business model, drop the country, abandon the Core Job — and only the founder, the CPO, or the chief of staff has the authority to run those moves and to override the bias at every level below them. Subtraction lives, when it lives at all, at the top of the org chart.
Wes Says This Only Works With a Trillion Dollars
The Asymptote Is a Product the Customer Never Has to Use
The limit case the catalog approaches but never reaches is the product that is never present at all. You wake up and the apartment is clean, with no cleaning product you remember touching. You arrive in another city for a meeting and no trip happened, because a Zoom call replaced what a decade ago meant an airport and a hotel. You file your federal taxes and there was no tax product anywhere in the experience. The Job was performed, and the artifact disappeared.
Altshuller's TRIZ tradition, working from engineering theory in the 1950s, arrived at the same idea from the other side: the Ideal Final Result is the function performed with no system left to perform it (see Scientific Foundations §20). Two independent traditions converged on one limit case.
The compass cuts both ways, and the pull toward the limit has a measurable shape. The customer's brain weights a removal roughly twice as heavily as an equivalent addition. The effect shows up most strongly with frictions, anxieties, chain-breaks, and switching costs — the largest class of removal targets. The brain registers the loss of one small Job at roughly twice the value of a new feature at the same cost (see Value Creation §7 and the loss-aversion treatment in Scientific Foundations §12).
Square's tip screen works because it removed the awkwardness of choosing an amount while the barista watches. TurboTax earned its position by removing the anxiety of am I doing this wrong; will the IRS catch a mistake. Each collected roughly twice the value an additive competitor could have bought with the same engineering budget. The asymptote pulls, and products that move toward it land harder than products that pile features against it.
A Note for the Reader Who Got This Far
You opened this book because something in your product was not landing the way you thought it would, and you have now spent a few weeks inside it. The substrate of the work is in your head: the brain as an investor of its own energy, the Job as the unit of motivation, the Graph as the structure the customer walks. The chain is that structure laid out in time, and the prediction error is the signal the brain produces when reality beats or misses the bar. A segment is a set of Graphs gathered around shared Core Jobs, and a mechanic is the move that bends one of them toward something more energy-efficient than what came before.
Underneath all of it sits one verb: subtract.
Subtract the Jobs the customer was performing only because no one had thought to remove them. Subtract the segments the team was serving because no one had run the margin, the assumptions the product was carrying because no one had tested them, and the customers whose cost-to-serve eats the margin earned on the rest.
Each subtraction hands something back — attention, time, energy, runway, focus, money. The customer's brain registers it at twice the weight of an addition, the income statement shows it on the next close, and the team's calendar feels it the following Monday.
What this book did not cover is the rest of the operation: how subtraction runs across all four pillars at once, the team-level resistance against it, the founder-level authority required to install it, and the diagnostic question that travels across the pillars — what has this team explicitly removed in each pillar in the last quarter, and why? It also left out why corporate product portfolios accumulate zombie products faster than startup portfolios do. That treatment belongs to a larger methodology, and to a book I have not finished writing yet.
For this one, you have enough. The compass is the removal of Jobs from the customer's day. The North Star is the product that was never present in the first place. Climb when you can. When you can't yet, ask one question at the next review: what would it look like for the customer to reach this outcome with no product there to perform it at all? Listen for the answers that feel embarrassing to give — those are the ones the operation is already pointing at.