Designing the Work

A vehicle moving through an armoring line passes through seven manufacturing disciplines: raw material, fabrication, welding, electrical, trim, upholstery, and paint and body. Every one of them can do excellent work and the vehicle can still come out late, wrong, or over cost, because the thing that determines the outcome is not the quality of any single discipline. It is whether each one knew precisely what it was receiving, precisely what it owed the next one, and precisely by when.

At Texas Armoring I ran that operation as acting chief executive, sixty-plus people directly and a worldwide network of more than fifteen hundred commission-based resellers behind them. The thing that actually held it together was not culture, and I want to be blunt about that because it runs against most of what gets written on this subject. It was a written, cross-departmental service-standards system that defined what marketing owed sales, what sales owed production management, and what production management owed each of those seven disciplines. Deliverables, timeframes, and accountability, on paper.

Most companies try to fix with culture what they created with design. It never works, and I think it is worth explaining exactly why not, because the failure is not a failure of sincerity.

Section 01Ambiguity is not neutral

When a role is unclear, the work does not stop. It gets allocated by a different mechanism, and that mechanism is social. In the absence of a written standard, the person who gets heard is the one who is most confident in a meeting, most comfortable interrupting, most fluent at reading a room, and most willing to advocate for their own contribution. Those are real skills. They are not the same skills as doing the work well, and there is no reason to assume they correlate.

So ambiguity is not the absence of a system. It is a system, and it is one that quietly selects for social fluency over output. It disadvantages introverts, people early in their careers, people operating in a second language, people from cultures with different norms around deference, and, in a way I have come to think is badly underappreciated, a very large group of people whose brains simply do not run on inference.

Chip Heath and Nancy Staudenmayer showed that leaders overwhelmingly focus their design attention on dividing work and almost none on reintegrating it, a bias they named partition focus.¹ The human-capital version of that same blind spot is that we write job descriptions describing a box and almost never describing the seams: what this person receives, from whom, in what condition, and what they are accountable for handing on. The box is the easy part. The seams are where people actually succeed or fail, and we leave them to be discovered.

Section 02What an internal service standard is, precisely

People hear the phrase and think bureaucracy. They are nearly opposites. A standard is a promise with a deadline and a named owner. Bureaucracy is a process with neither.

A usable standard has four parts, and if any one is missing you have written a preference rather than a standard:

• The deliverable, defined by what the receiving function needs, not by what the sending function finds convenient to produce. This is the part everyone gets backwards.
• The timeframe, expressed as hours or days from a triggering event, never as “at the weekly meeting.” A standard tied to a calendar is a standard with a built-in delay of up to seven days.
• The acceptance criterion: the one thing the receiver checks to know it is done. Without this, “done” becomes a negotiation, and negotiations cost more than the work they are about.
• The escalation path: a named human, not a queue. A standard without an escalation path will be quietly broken within a quarter, and nobody will report it, because there is nobody to report it to.

That system did not make Texas Armoring a more pleasant place to work in some abstract way. It made it faster, because when the interface is written down, the daily cost of crossing it collapses toward zero. Nobody renegotiates the terms of a handoff every time a handoff occurs. And it made it a more survivable place to work for people who had previously been spending a substantial share of their cognitive budget just trying to determine what was expected of them.

Section 03The population nobody designed for

Deloitte estimates that somewhere around one in five adults is neurodivergent, and that disclosure rates are low enough that the real figure inside any given workforce is probably higher.² That is not a niche. Every company above fifty people already employs a meaningful number of neurodivergent people, whether or not it knows, whether or not it has a program, and whether or not anyone has ever said so out loud.

The numbers coming out of structured hiring programs are striking. JPMorgan Chase’s Autism at Work initiative reported that participants in its first cohort were around forty-eight percent faster and up to ninety-two percent more productive than neurotypical peers in comparable roles. Retention across the major corporate programs at Microsoft, SAP, JPMorgan Chase, and EY consistently runs above ninety percent. Deloitte has reported that teams including neurodivergent professionals can be up to thirty percent more productive. And the Job Accommodation Network finds that roughly six in ten workplace accommodations cost nothing at all.³

I am going to argue against the popular reading of those numbers, because I think the popular reading is both wrong and less useful than the truth.

These figures come from opt-in programs, in deliberately selected roles, with structured onboarding, trained managers, and explicit support. That is a substantial selection effect and anyone citing the numbers without saying so is doing advocacy rather than analysis. The honest conclusion is not that neurodivergent people are inherently more productive than everyone else. It is something considerably more interesting and considerably more actionable:

Section 04The curb-cut effect, in an operating context

Curb cuts were built for wheelchairs. They are used overwhelmingly by people with strollers, delivery carts, luggage, bicycles, and knees that hurt. The accommodation turned out to be a general improvement that nobody had bothered to design because nobody had been forced to.⁴

Every single thing on the list below was introduced somewhere as an accommodation, and every one of them made the whole organization measurably faster:

• Written agendas circulated before the meeting. Framed as support for people who process better with preparation time. Actually eliminates the meeting where the first twenty minutes are spent establishing what the meeting is about.
• Explicit acceptance criteria on every deliverable. Framed as support for people who find implicit standards difficult to read. Actually eliminates rework, which is the single largest hidden cost in knowledge work.
• Interview processes with no surprises: the questions in advance, the format described, the assessment tied to the actual job. Framed as accommodation. Actually a better predictor of performance, because you are now measuring the work instead of measuring who is calm under artificial pressure.
• Decisions recorded asynchronously, with the reasoning attached. Framed as support for people who cannot process a fast verbal debate in real time. Actually the only way an organization retains institutional memory past the tenure of the people who were in the room.
• Direct feedback rather than hinted feedback. Framed as support for people who do not decode subtext reliably. Actually the difference between a person who can fix a problem and a person who has been anxious for three weeks about something nobody would name.

None of those are kindnesses. They are operating improvements that happened to arrive through the door marked inclusion, and the fact that they arrived through that door is a fairly damning comment on how the rest of us have been designing work.

Section 05Hire for decisions, not adjectives

Most job descriptions are lists of adjectives: strategic, detail-oriented, collaborative, self-starting. Nobody has ever failed a role because they were insufficiently collaborative in the abstract. People fail roles because they could not make a specific class of decision well enough, fast enough, or often enough.

So rewrite the description as a list of the decisions the person will own. “This person decides which four products we launch and in what order.” “This person decides when a batch ships and when it does not.” “This person decides what we tell a customer when we are going to be late.” The exercise is uncomfortable, which is the point, because two things immediately surface. First, you will find decisions that nobody currently owns, which are exactly the ones that have been taking six weeks. Second, you will find decisions that three people own, which are exactly the ones that generate the recurring argument everyone has stopped mentioning.

You will also interview differently, because you now have something real to test. And you will hire differently, because a person’s track record of making a specific class of decision is observable, whereas “strategic” is a vibe.

Section 06What AI actually does to human capital

The prevailing anxiety is that artificial intelligence eliminates human capital. What I have observed at close range is different and more consequential. It eliminates the assembly work that was concealing the judgment work.

A large fraction of most knowledge jobs was never judgment. It was locating the current number, reconciling two systems that disagreed, formatting the thing, remembering what was decided in March. When that fraction gets compressed, what remains is the part that requires a person to decide something under uncertainty and be accountable for it. That raises the value of judgment sharply, and it raises the cost of ambiguity by exactly the same amount, because ambiguity is now consuming a much larger share of a much more expensive activity.

There is a second-order effect that I have not seen many people name. These systems can only work with what has been written down. If your organization’s real operating knowledge lives in the heads of three long-tenured people and has never been documented, no model can help you, because there is nothing for it to read. The companies getting genuine leverage are, almost without exception, the companies that had already done the unglamorous work of making their processes explicit. The technology did not reward the most sophisticated organizations. It rewarded the most legible ones.

Section 07The part that is not about efficiency

I have made an efficiency argument for six sections because efficiency arguments are the ones that survive budget season. But the reason I care about this is not efficiency, and I would rather say so plainly than pretend otherwise.

People rarely leave a job because the work is hard. In my experience they leave because they cannot tell whether they are doing it well. That is a specific and corrosive kind of suffering, and it is almost entirely manufactured by the organization rather than by the work. A person who knows exactly what is expected, who receives direct feedback, and who can see how their output connects to something real will tolerate an extraordinary amount of difficulty. A person who is performing well but cannot confirm it will burn out doing considerably less.

Every leader has the capacity to remove that particular suffering, and it costs nothing except the discomfort of having to decide what you actually want, in writing, where you can be held to it. I think that discomfort is the real reason it does not happen more often. Vagueness protects the person setting the expectation, not the person meeting it.

The single highest-leverage thing I have ever done for the people who worked for me was not a values statement, an offsite, or a survey. It was writing down what “done” meant, and then holding myself to it first.