The Great Role Blur

For years, the conversation about artificial intelligence has circled one anxious question: will it take our jobs? It is an understandable fear, and a strangely small one. It imagines the shape of work staying fixed while the workers are swapped out, one human for one machine, the org chart untouched. But that is not what is happening. Something larger and less tidy is underway.

The better question is this. What happens when AI does not replace the people inside an organization, but the assumptions the organization was built on?

Because every company is, underneath its product and its logo, a set of beliefs about how work should be divided, measured, and moved from one hand to the next. Those beliefs hardened into departments, titles, reporting lines, and the slow choreography of meetings. We stopped seeing them as choices and started treating them as physics. AI is now pressing on those beliefs, and the structures we mistook for permanent are turning out to be assumptions all along.

The implications run far deeper than automation.

Smaller Teams, Bigger Output

Historically, growth meant hiring. More products required more engineers. More engineers required more managers. More managers required more coordination, which required more people to manage the coordinating. Scale was something you bought by the head.

AI breaks that arithmetic. For the first time in modern memory, a company can grow its output without growing its headcount at the same pace. A developer with capable tools can produce what once took several. A designer can explore dozens of directions in an afternoon that would have filled a month. A product manager can read across a mountain of research and feedback in the time it used to take to schedule the meeting about it. As the models improve, that multiplier compounds.

But the multiplier is lumpier than it sounds. It is enormous at the edges of the work, the boilerplate, the first draft, the thing that has been built a hundred times before. It is far smaller at the center, where the real labor is rarely producing something new. It is understanding something old well enough to change it safely. Most work is not a blank page. It is a system someone else built, and the slow, careful reading of it before you dare to touch anything. AI is dazzling at the page and still clumsy at the system. So the gains arrive unevenly, and the parts that look most automatable from a distance are often the ones that resist it most up close.

Even so, the conclusion looks obvious. Companies will need fewer people to make the same amount of work. Not zero people. Not fully autonomous firms humming along in the dark. Just fewer.

But here I want to slow down, because the obvious conclusion hides a long-running pattern that often points the other way. When something becomes dramatically cheaper to produce, we rarely keep buying the same amount of it and pocket the savings. We buy more. Cheaper cloth did not shrink the clothing trade. Cheaper computing did not end the demand for software, it created an appetite for software no one had imagined. Economists call the instinct to assume a fixed amount of work the lump of labor fallacy, and it has been wrong about every major technology for two hundred years.

So the honest version of the claim is subtler. Fewer people per unit of work does not cleanly mean fewer people. It may mean the unit of work stops being the thing worth counting. The question is no longer how many hands a task requires, but how much ambition a company can suddenly afford. Some organizations will use that surplus to do the same with less. The interesting ones will use it to attempt things that were never on the table before.

Both futures are real. Which one a company chooses tells you almost everything about what it values.

The Rise of the Reviewer Economy

The most profound shift may be the move from making to checking.

Picture a software organization a few years out. AI drafts most of the code. AI writes the first version of the documentation. AI generates the test plans, the design variations, the research summaries. The human role narrows to a single, powerful verb: approve. People become validators, directing and correcting and signing off on work that machines produced.

On paper, this is pure efficiency. In practice, it carries a risk we are not yet talking about honestly.

There is a quieter danger beneath the obvious one. Reviewing the work of a machine is harder than reviewing the work of a person, and harder in a way that hides itself. When a colleague hands you something flawed, you can ask what they were thinking, and the answer teaches you both. The machine has no thinking to interrogate. It offers you something plausible, fluent, and confident, with no story behind it and no one to question. And plausible is exactly the thing a tired reviewer waves through. The failure mode of the reviewer economy is not the dramatic mistake. It is the rubber stamp. The assumption nobody chose, carried forward because it looked finished.

Humans develop mastery by doing, not by reviewing. The judgment that lets a senior engineer glance at a system and feel where it will break was earned by building systems that broke. The instinct that tells a designer a flow is wrong before she can explain why was paid for in years of making flows that were wrong. Expertise is not downloaded. It is accumulated, slowly, through contact with the actual material.

If a generation spends its formative years approving machine output rather than wrestling with the work itself, where does the next layer of mastery come from? The reviewer economy may buy a great deal of short-term speed while spending down a reserve we did not know we were drawing on. Efficiency now, capability later, and the bill arrives in a decade when we look around for the people who were supposed to have become masters and find a generation of skilled approvers instead.

The End of Specialized Silos

For decades, organizations were built around specialization. Engineering. Design. Research. Product. Operations. Each a discipline, each a territory, each with its own language and its own walls.

AI does not respect those walls. A designer can now produce working code, enough to make a real thing run, even if not yet enough to survive five years in production with six other systems leaning on it. An engineer can sketch a credible interface. A product manager can stand up a working prototype before lunch. A researcher can synthesize a study automatically. The boundaries that organized the modern company were never laws of nature. They were a way of coping with the fact that no single person could do everything. As that constraint loosens, the boundaries loosen with it.

Companies have a polite phrase for this. They call it role blurring. But blurring undersells it. What is actually happening is capability convergence. Work is detaching from the department it used to live in and reattaching to the outcome it is meant to serve. The question shifts from what is your function to what can you actually move.

The people who thrive in this will not be the deepest specialists, narrow and excellent in one lane. Nor will they be shallow generalists who skim every surface and master none. They will be the ones who can range across several domains and still know where human judgment has to enter, the places where the machine is fluent but not wise.

The Management Collapse

A different shift is moving through organizations more and it touches the middle.

Management layers grew, in large part, because information traveled slowly. Someone had to gather updates, route tasks, consolidate what was happening on the ground, and carry it upward and back down again. Managers were, among other things, the nervous system of a company in an era when signal moved at the speed of a status meeting.

AI performs a striking amount of that work now. Dashboards report in real time. Tasks orchestrate themselves. Knowledge sits a question away rather than three forwarded emails deep. And so the uncomfortable question surfaces in leadership meetings everywhere. Do we need this many managers? Many companies are already widening spans of control, asking a single manager to hold twenty or thirty people where ten was once the ceiling. The pyramid begins to flatten.

But here, too, I want to resist the clean story. It is tempting to say management was only ever coordination, and that coordination is now automated, so management dissolves. That is half true and dangerously half. Coordination was always the visible part of the job. The invisible part was absorbing ambiguity, holding accountability when something failed, shielding a team from chaos so it could think, noticing the engineer who has gone quiet, deciding which fire actually matters. A dashboard automates the reporting. It does not automate the care, or the judgment, or the weight of being the person responsible when the plan meets reality.

So the management layer will not so much collapse as transform. The manager as task router and update collector is genuinely fading. The manager as coach, as steward of judgment, as the human who decides what is worth doing and protects the conditions for doing it well, becomes more necessary, not less. The pyramid flattens. What it flattens into depends entirely on whether companies remember what managers were carrying all along.

The Disappearing Entry-Level Job

The hardest problem sits at the bottom of the ladder, and it is the one I think about most.

Junior roles were never only about the work juniors produced. They were apprenticeships in disguise. You started with the simple tasks, the ones nobody senior wanted, and in doing them badly and then less badly you learned the texture of the craft. The simple work was tuition. You paid it in small assignments and were repaid in judgment.

AI is absorbing exactly those tasks first, because they are the most automatable. And this creates a paradox with no easy way out. Companies may genuinely need fewer juniors to get the work done. But every senior employee a company will ever have was once a junior who learned by doing the work that is now being automated away. Remove the bottom rung and the ladder still stands, for now, held up by people who climbed it before it disappeared. The trouble comes for the generation that arrives to find the first rung gone, and the one after that, who were meant to be trained by people who never got to climb.

This may become the defining labor question of the next decade. Not how do we replace workers, but how do we still grow them. An apprenticeship crisis is a slow emergency. It does not announce itself. It shows up years later as an absence, a missing layer of mastery that no one decided to remove. We simply optimized the entry level until it was gone.

Why UX Should Be Paying Attention

Design organizations are unusually prone to misreading this moment. People watch AI generate wireframes and conclude that UX is dissolving. I believe the opposite, and I believe it for a specific reason.

What is becoming cheap is interface production. Pixels, layouts, variations, the mechanical surface of the work. What is becoming scarce, and therefore precious, is human understanding. As the cost of making a screen falls toward zero, the value migrates to the questions the screen was supposed to answer. Should this exist. What does a person actually need here. What will this do to them that no one intended.

So the designer of the near future spends less of her day moving rectangles and more of it doing the things a machine cannot. Understanding human behavior in its full, contradictory texture. Shaping product strategy. Naming the unintended consequence before it ships. Facilitating the decision in the room. Building trust, which is slow and human and cannot be generated. Designing the collaboration between people and AI so that it amplifies rather than erodes.

The mechanical part of design is being automated. The human part is becoming the entire point. This is not the end of the discipline. It is the discipline finally being asked to be what it always claimed to be.

The Future Belongs to Judgment

Gather these threads and they pull in one direction.

The winners of this era will not be the fastest producers. AI is becoming the producer, tireless and cheap and endlessly available. When generation is abundant, the scarce thing is discernment. Knowing what to build, and the rarer wisdom of knowing what not to. Knowing when the machine is confidently wrong. Knowing what people actually need underneath what they ask for. As output becomes free, the ability to judge it well becomes the whole game.

This is why the organizations that endure will not be the ones that automate the most. Automation is easy to buy and easy to copy. The harder, rarer thing is knowing where human wisdom still has to sit, and building the company so that wisdom is grown rather than spent.

Because the real story was never humans against AI. It was always humans learning to supervise systems that increasingly do the work themselves. And inside that arrangement lies a choice we keep pretending is a forecast. We can design organizations that use the surplus AI creates to enlarge what people are capable of, to free them toward judgment and meaning and the work only they can do. Or we can design organizations that reduce people to reviewers of machine output, efficient and idle and slowly forgetting how to make anything themselves.

That choice, far more than any breakthrough in any model, will decide what the future of work becomes. It is not being made in a lab. It is being made, in every company that decides what to do with its sudden abundance. And it is still, for now, ours to make.