A Failure of Imagination · Nicholas Rinard

We may be using AI to make one of the most impersonal features of modern life almost infinitely scalable: the requirement that people simplify themselves until a system can understand them.

The shoes are the right length, but they rub your heels raw. On the return form, you choose “too small” because it is the least-wrong option in the dropdown. At work, your project-management app offers “to do,” “doing,” or “done.” You have finished your part but are waiting for someone’s approval, so the task stays “doing.”

The return records a size problem. The task board records unfinished work. Neither quite describes what happened, but you have to use the choices available. Now imagine being able to show someone exactly where the fit is wrong.

A gardener brings a trowel to a craftsman’s bench. The handle presses into the gardener’s palm, and after an hour of planting, the hand aches. Nothing is broken. Buying another of the same design would leave the same problem. The craftsman watches the gardener grip it and begins shaping a replacement.

The gardener tries the new handle. There is still an awkward spot beneath the thumb, another where the palm rests. The craftsman marks those places, removes a little wood, and hands it back. Later, in the garden, the gardener might work for a while before noticing they have stopped shifting their grip. A detail the craftsman cared about has become useful in the gardener’s hands.

A gardener tests a custom wooden tool handle while a craftsman marks pressure points, with standardized handles behind them and a coral line leading to comfortable use in the garden.

The gardener’s grip guides the next adjustment.

Each adjustment takes time, and the tools on the bench affect which shapes the craftsman can reasonably make. A simpler handle may be the only one the gardener can afford. Making the same shape repeatedly brings the price down further, putting useful tools into many more hands. But the places where that shape does not fit are left for each gardener to accommodate.

Over time, that practical compromise can become an assumption no one revisits. The craftsman stops offering shapes that take too long to make. The gardener stops asking whether the handle could fit better. We do the same at work when we stop explaining why a task does not fit the board and learn to keep track of the difference somewhere else.

Then AI arrived, and almost instinctively we asked the question those systems had trained us to ask: How much more can we get done?

More emails, more ads, more candidates, more content, more code.

But what could we make possible that we had stopped trying to do?

If AI makes it cheaper to care more, then using it merely to do more of the same, faster, is a failure of imagination.

The Checklist Factory

Making more of the same gets easier when the work follows a repeatable sequence. In a workshop producing thousands of handles, one person can cut the blanks, another shape them, and another check the dimensions. The craftsman’s knowledge becomes a specification others can follow. That makes useful work more dependable and affordable, without requiring everyone to learn it all from the beginning.

But the specification can tell the last person whether a handle matches the drawing; it cannot tell them whether the gardener’s hand still aches. Answering that requires someone to follow the work beyond their own station and have the authority to change what the others produce. As the work is divided, that responsibility can disappear even while every assigned task is completed.

I call this the checklist factory. Finishing your step and passing the work along becomes the outcome. Everyone owns a piece; no one remains responsible for what those pieces actually do for the person they were meant to serve. In an office, the stations might be sales, implementation, and support. Each team can fulfill its commitments while the customer is still struggling with the problem that brought them there. The organization can verify that the work happened more easily than it can establish whether the work helped.

For the person inside it, following the path becomes safer than questioning where it leads. Ownership ends at the handoff. Someone may see exactly what is wrong and still lack the time or authority to fix it. Netflix’s original culture deck traced how an organization can come to reward that obedience.

A company grows, and its work becomes more complex. As it hires, the proportion of exceptional performers falls. Eventually, the complexity exceeds what the people can manage informally. Errors multiply. Management introduces procedures to bring the chaos under control, and the relief is real: fewer mistakes, more predictable work. But the growing emphasis on following procedures drives more of the independent, inventive people away.¹

The company can then become remarkably successful at its existing business. It runs efficiently, makes few mistakes, and may lead its market. That success conceals what it has lost: fewer people remain who will question the way the work is done.

Then a new technology, competitor, or business model changes the market. The employees have become excellent at following the old processes, and adherence is what the organization rewards. The very habits that made it efficient now make it slow to adapt. It can keep executing correctly as its business becomes irrelevant.¹

Diagram showing customer needs changing while system output continues unchanged; internal metrics stay green and more process reduces the ability to adapt.

The dashboards can stay green after the destination has moved.

For the people working inside that arrangement, the loss of judgment has another consequence: it makes the fear of being replaced by AI easier to understand. If success means completing your assigned step and passing the work along, a system capable of doing that step faster looks like a substitute for you. The knowledge you could bring to the whole outcome has already been pushed outside the job.

Personalized for Which Outcome?

Giving people more room for judgment might seem unnecessary if a system could understand each person well enough. A social feed shows why that is only part of the answer. It learns what makes you pause and keeps you looking after you meant to stop. You put the phone down after spending longer than you intended, with little sense that the time was worthwhile. The feed can be unmistakably yours without feeling as though it was made for you.

In a study of 806 Twitter users, an engagement-ranked feed amplified more politically hostile content than a chronological one, even though participants did not prefer the political posts it selected. The study examined a snapshot of posts, rather than long-term effects.²

Here, the checklist factory has learned to vary its output for every individual. Understanding what produces a response is still different from taking responsibility for whether the experience serves the person. Personalization changes what comes off the line; it does not decide where the line should lead.

A man sets down his phone while a personalized-feed machine turns the constellation of his life into an endless ribbon optimized for attention.

Personalized for which outcome? A system can know what holds your attention without asking what deserves it.

Define Excellence First

Once a process is running, its results give us plenty to improve. We can make a feed more engaging or a hiring process more selective without revisiting the assumptions that gave the work its shape. A recruiting team can rigorously assess every applicant and still know nothing about the person it never thought to approach.

I began questioning that starting point after listening to Adam Ward discuss recruiting at Cursor on Lenny’s Podcast. He described the recruiting “funnel of doom”: contact a broad population, work with whoever responds, filter through interviews, and hire from whoever remains. The people who survive may be the strongest of those who replied, while the people the role actually calls for remain outside the search.³

His alternative is a “pillar of excellence.” Before opening a search, define what exceptional performance would mean in this particular role. Identify the evidence that would distinguish someone who can do that work. Then map the people who might meet the standard and pursue them deliberately.³ The ambition begins with the outcome, before the available inputs narrow it. The search can also teach us that we have misunderstood the work itself.

What interested me was how that attention could continue once someone agreed to talk. Adam described choosing a candidate’s next interviewer based on something learned in an earlier conversation, rather than simply whoever happened to be available.⁴ Imagine arriving and realizing that something you said last time helped shape the conversation you are about to have. You do not have to begin again. What the company learned about you has changed what it does.

That is a different kind of consistency: the seriousness of the assessment holds while the path responds to the person.

A hand places a coral standard at the center of a map, with distinct paths connecting different people to the same high bar.

The standard is common. The paths are not.

Adam’s claim is about recruiting. He distinguishes it from sales and says explicitly that not every funnel is bad.⁴ The broader question is mine. Once we can describe what excellent work requires, how much of its apparent cost comes from trying to fit it into a process built for something else?

A deliberately mapped search may redirect effort that would otherwise go into assessing people who were never likely to fit. But sustained attention still requires research, preparation, coordination, and memory. It has to survive the next conversation and the next handoff. That is the supporting work whose cost we now have reason to reconsider.

The Cost of Looking Again

Following a customer’s need through those handoffs takes more work than passing along the request. Imagine a customer asking to export order records into a spreadsheet. Support records the request, a product manager puts it on the roadmap, and an engineer builds the button. Each person has done their part. But no one has gone back to ask what happens after the customer downloads the spreadsheet.

Finding out means revisiting old conversations, comparing them with support requests, and watching how the customer actually works. Those hours compete with the features the team has already promised. The export button is a clear, bounded job; questioning the workflow behind it opens up work no one has budgeted for.

This is where AI can change the calculation. It can help bring the scattered history together so the team spends less time locating and assembling the evidence. People still need to check that account with the customer and pursue what the records leave out. But an investigation that kept losing out to the next ticket may now be practical.

Suppose the team learns that someone exports the orders each evening to compare them with invoices in another system. Records disagree, and the employee has to track down the differences by hand. A faster export leaves that evening’s work in place. Repairing the connection between the systems could remove much of it. What began as a request for a button becomes a reason to reconsider why the customer has to reconcile the records at all.

The team now has to decide whether to take responsibility for fixing that recurring problem. Repairing the connection requires testing, maintenance, and changing work already on the roadmap. It has to be worth that cost. Fewer recurring support problems or a customer who stays and recommends the product may provide a return. Some of the value will belong to the customer without coming back to the company, and the company still has to decide how much of that work it can support.

If the change works, the employee can check that the records agree, close the laptop, and get the evening back. That is also an experience a competitor can offer. Where customers can recognize the difference and switch, the company that declined the work because of its cost must reckon with the cost of losing them. What once seemed an acceptable burden can become a reason to leave.

The opportunity with AI is not just to automate what we already do. It is to revisit the compromises we made when attention was expensive.

The export button was the request the customer knew how to make. Understanding the evening behind it gave the team a different purpose. People have been making that kind of discovery long before these tools existed—and the work can change when they are allowed to act on it.

What Craft Gives Back

During a busy lunch at Eleven Madison Park, Will Guidara overheard four guests reflecting on a week spent eating at some of New York’s most celebrated restaurants. They were leaving for the airport after lunch. The only thing they regretted was never trying a New York hot dog.

Guidara slipped out of the restaurant, bought one from a street cart, and persuaded the chef to serve it. The kitchen cut it into four pieces and plated it with the same precision it brought to the rest of the meal. Before the guests received their final savory course—a duck the restaurant had spent years perfecting—they received the hot dog they had never asked for.

Guidara had served more technically accomplished and far more expensive food. He had never seen a table respond with such joy.⁵

The hot dog was not a rejection of standards. It mattered because the standards around it were so high. But the restaurant’s deepest promise was not that every guest would receive precisely the same sequence of perfect dishes. It was that every guest would leave having experienced something extraordinary. For those four people, fulfilling that promise required changing the sequence.

Guidara later created a role called the Dreamweaver to help the staff turn what they noticed into experiences they could deliver. The restaurant did not write a checklist for when to buy a hot dog. It made noticing consequential.⁵

The guests felt that someone had paid attention to them. The people serving them felt the pleasure of doing more than executing a service correctly. They could use judgment, imagination, and everything they had learned about the people in front of them to make the experience their own.

The Work Behind the Task

That feeling is easy to dismiss as a luxury of fine dining. But the same architecture can restore something more fundamental.

Dutch home care had been divided into authorized services performed by different workers. A nurse might handle the task requiring clinical expertise while others handled bathing, dressing, or daily needs. Each service could be completed correctly while patients experienced a revolving sequence of strangers and no one retained the meaning of the whole.

Jos de Blok, himself a nurse, created Buurtzorg in response. Small, self-managing teams became responsible for the full course of a person’s home care. They assessed needs, changed care plans, coordinated with families and physicians, and worked toward helping each person regain as much independence as possible. An information system and a small back office supported the teams’ administrative work, helping the nurses coordinate care.

A Buurtzorg nurse named Deborah Warta recalled caring for a young man who was paralyzed. Her team came each morning to provide the care he needed. His wife told them that their presence allowed her to begin the day as his partner instead of as his nurse.

The completed tasks mattered. But what the work gave this couple was part of their normal life together.⁶

It gave something back to the nurses as well. They could use the full range of their abilities, respond as circumstances changed, and see what their work made possible. The supporting system did not make the nurse more interchangeable. It made it easier for the nurse to practice the craft of nursing.

A Michelin-starred dining room and a young couple receiving care at home could hardly look more different. What connects them is not extravagance. In both, a person close to the work understood the human outcome, and what they understood was allowed to change what happened next.

In one scene a server departs from a tasting menu to present four guests with a carefully plated hot dog; in another a nurse adapts care so a couple can share an ordinary morning as partners.

Craft changes the experience on both sides of the work.

Standardize the High Bar

Consistency no longer has to mean giving everyone the same process. It can mean taking every person’s outcome equally seriously.

Standardize the high bar, not the path.

Guidara held to a promise of an extraordinary experience while the meal changed. Buurtzorg pursued independence and continuity while nurses planned with each person. Shared standards gave the work direction, and people close to the outcome had room to act on what they learned. Specialists could still contribute their parts, with someone responsible for keeping the purpose of the work intact across the handoffs.

The Feeling of Craft

Some of the care in an object may be invisible to the person who receives it.

A cabinetmaker may smooth and finish the back of a cabinet that will sit against a wall. Most people will never see it. But the maker handles the piece before it is installed, feels the roughness, and knows where the finish stops. Another pass brings the object closer to what they intended to make.

That does not mean pursuing perfection without regard for purpose. Craft is the judgment to know which details preserve the integrity of the whole—and the willingness to address them after “acceptable” offers a convenient place to stop.

My wife, Charley, has a favorite line she made up herself: “If a little effort goes a long way, imagine what it’d be like if you actually tried.”

Trying is not only expenditure. For people who care about their work, it is one of the ways they recognize themselves in it. The final pass, the thoughtful adjustment, the detail no one demanded—these are not just improvements to the output. They are evidence that a human being was fully present in making it.

An organization can give that care somewhere to go. The question is whether it can keep doing so after the exceptional encounter is over.

Craftsmanship at Scale

I think the possibility now opening deserves a name: craftsmanship at scale. It means making the experience at the heart of craft available to more people: feeling seen in what you receive, and being trusted to make something worth standing behind.

The words “at scale” carry an obligation. The next person through the door should have reason to expect that same seriousness, even when someone else is working. What one person learns should help the next person do better work. Shared tools and accumulated experience allow that understanding to grow beyond a single relationship. Each outcome can receive attention without the work beginning from zero.

Different workers and people gather around a shared constellation of accumulated understanding while a craftsperson shapes a wooden joint at its center.

The next person inherits the understanding, not the answer.

Craftsmanship cannot depend on heroism. If a nurse, engineer, or server must fight the organization to do excellent work, the institution is consuming that person’s care as a private subsidy. Over time, they may tire, leave, or learn to stop offering what the job seems determined to refuse. Making room for craft means giving them time to follow a concern, authority to act on what they discover, and colleagues who can carry the work forward.

The promise is not that everyone will try harder. It is that trying can stop feeling like resistance.

For the person receiving the work, that means being understood can become something they expect from the place itself. Less depends on their luck in finding the one person willing to fight for them. We began with people simplifying themselves until a system could understand them. Craftsmanship at scale asks the system to do more of the adapting.

What We Choose to Make Possible

None of this happens automatically. Lowering the cost of attention does not make an institution attentive. A system built to reward quantity can turn every hour recovered into another demand for output.

The choice becomes real when someone discovers something that would be easier to ignore. A team learns that serving the customer well means reconsidering work it has already committed to deliver. The person raising the concern has understood more and taken responsibility for the result. What happens next teaches everyone watching how much that judgment is actually worth. We cannot promise ownership of the outcome while making it unsafe to question the work.

Somewhere in work you know well, there may be a compromise you have learned to stop mentioning: a detail that keeps being missed, a person the usual answer never quite serves, something you once hoped to make better. You may still be working around a limit that has loosened since you last tried. What would it take to give that another serious attempt? Sometimes the answer begins with making room for someone else’s judgment to reach us.

There will still be reasons to stop. Time and money remain finite, and a thoughtful person can decide that the ordinary answer is the right one. Craft includes knowing where another pass will matter. We can reopen compromises whose justification has changed without pretending that every constraint has disappeared.

People have cared deeply about their work throughout this history. They have found ways to make something excellent inside constraints that gave them little room. I want more of that desire to survive an ordinary working day. More people should be able to use what they know, bring their taste and imagination to the work, and see it make a difference to someone.