In the Age of AI, Responsibility Lives in the Approval Button

I’m not sure when I first watchedThe Jetsons. I was born in 1967, so it was probably sometime in the late 1970s or early 1980s. It was Hanna-Barbera’s futuristic answer toThe Flintstones, and like a lot of kids back then, I loved it. My friends and I couldn’t get enough of science fiction. Robots, space, monsters, the future—we were fascinated by all of it.

One of the running jokes in The Jetsons was George Jetson’s job. Every day he went to work, sat in front of a giant control panel, and spent his day pushing buttons while machines did everything else. He was a master button pusher. The joke was simple: anyone can push a button. Somehow, technology had reduced human labor to sitting in a chair and pressing “Go.”

For most of my life, I thought it was just a funny cartoon. Then, a few months ago, I was working with Claude Code.

The AI was happily writing code when it suddenly stopped and asked for permission to perform an action. I clicked Allow. It went back to work for another minute, stopped again, and asked for permission. I clicked Allow again. Then it happened again… and again… and again.

After the tenth or twentieth time, I remember thinking, “Why am I even here? Why can’t the AI just do everything?”

Then it hit me.

Maybe this wasn’t an annoying limitation. Maybe this was what work was beginning to look like.

I had officially joined the growing number of people who spend part of their day clicking buttons labeled Approve, Accept, Publish, Deploy, Authorize, or Confirm. The buttons look simple, but the machinery behind them is becoming far more powerful than anything George Jetson ever imagined. He was operating machines. We’re beginning to operate artificial intelligence.

AI can already write a marketing campaign, generate software, analyze a market, prepare legal documents, recommend medical treatments, evaluate loan applicants, respond to customers, create financial forecasts, and perform countless other knowledge-based tasks. Increasingly, the human isn’t doing the work anymore. The human is reviewing what the machine has produced and deciding whether to let it move forward.

That is a profound shift in the nature of work. We are moving from doing the work to authorizing the work.

The Entrepreneur Was Always the Decision-Maker

In one sense, this is not entirely new. The world has always rewarded people who make good decisions. In business, we call these people entrepreneurs.

An entrepreneur sees an opportunity, develops a vision, decides where to invest time and money, and accepts the risk of being wrong. The entrepreneur is not necessarily the person performing every task. In fact, successful entrepreneurs rarely are. Their value comes from deciding what should be built, why it should exist, who it should serve, and how the pieces should fit together.

The problem has always been execution.

An entrepreneur may see the whole picture, but no single person possesses every skill needed to turn a vision into reality. To build anything meaningful, the entrepreneur has traditionally depended on other people: developers, designers, salespeople, accountants, marketers, lawyers, managers, bankers, technicians, and employees at every level.

This has always been one of the great challenges of business. The world does not have a shortage of intelligent people. What is much harder to create is teams or a company of coordinated intelligence directed toward one purpose.

When you put a group of capable people together to pursue someone else’s vision, something often gets lost. Information is misunderstood as it moves through the company. Departments develop their own priorities. Managers protect their positions. Employees focus on what they are measured on. Responsibility becomes divided among so many people that no one feels fully responsible for the final result.

The founder may begin with a clear idea, but by the time that idea has passed through meetings, project managers, designers, developers, legal reviews, budget discussions, and layers of approval, the final product may barely resemble the original vision.

That is not always because the people involved are unintelligent or lazy. It is because ownership changes behavior. The entrepreneur is risking capital, reputation, time, and often a large portion of their life. The employee is usually working on one part of that vision in exchange for a salary. For the entrepreneur, the mission may be deeply personal. For the employee, it may simply be a job.

There is nothing immoral about that. Most people work because they need to earn a living, not because they have adopted the owner’s dream as their own. But the difference in commitment, urgency, and emotional ownership is real, and it affects production.

Human beings are also inconsistent. We start businesses, diets, degrees, relationships, exercise programs, and personal projects with the best of intentions, yet many of them are abandoned. Intelligence does not guarantee discipline. Desire does not guarantee long-term commitment. Knowing what should be done is very different from continuing to do it when the work becomes boring, difficult, or uncertain.

Companies inherit all of this human inconsistency at scale. Every person added to an organization brings another set of incentives, distractions, fears, ambitions, interpretations, and limitations. The larger the company becomes, the more energy it must spend simply getting people to move in the same direction.

Hierarchy Was the Original Approval System

The traditional answer to this problem was hierarchy.

Established companies tend to manage coordination better because their pipelines have already been built. Responsibilities are divided, procedures are documented, authority is assigned, and employees are placed into defined roles. One person prepares the work, another checks it, another approves it, another releases the money, and another reports the outcome.

In that sense, hierarchy has always been an approval system. It allows a large organization to function without requiring every person to understand the entire business.

This creates stability, but it also creates friction. Every handoff introduces delay. Every management layer creates another opportunity for confusion. Every approval point makes it possible for someone to avoid risk, protect themselves from blame, or slow down a decision that no one feels personally responsible for making.

Large organizations compensate by adding meetings, policies, reporting systems, performance measurements, management structures, and software. These systems make cooperation possible, but they also consume an enormous amount of human energy.

The entrepreneur’s challenge has never been merely finding intelligent people. It has been turning many different people’s intelligence into dependable execution of a vision they do not personally own.

Artificial intelligence may change that old bargain.

Until now, an entrepreneur who wanted more production usually needed more people. More people required more management. More management required more procedures, meetings, communication, and oversight. Growth almost always meant building a larger hierarchy.

AI creates the possibility of scaling production without scaling the human organization at the same rate.

A small company may be able to use AI systems to research markets, develop campaigns, create software, prepare reports, monitor performance, communicate with customers, and coordinate other specialized agents. The founder still needs to decide what should be built, determine the goals, evaluate the results, and accept the risk, but far fewer people may be needed to translate those decisions into action.

In the old model, the entrepreneur made the decision and then depended on an organization to execute it. In the emerging model, the entrepreneur makes the decision, AI performs much of the execution, and the entrepreneur reviews and approves the result.

The bottleneck begins to move from labor and coordination to judgment and button pushing.

The Human at the Button

This is where the approval button becomes far more important than it first appears.

If an AI makes a terrible mistake, who is responsible?

You can’t fire it. You can’t sue it. You can’t send it to prison. You can’t ask it to explain itself to a grieving family or a board of directors. The AI may have generated the recommendation, but it cannot accept responsibility for the consequences.

This is why the phrase human in the loop ” matters. It describes systems where AI performs most or all of the work while a person remains responsible for the final decision.

As AI becomes more capable, that human role begins to shrink physically while growing morally.

An AI system might spend hours researching, coordinating specialized agents, testing solutions, evaluating risks, and preparing a recommendation. The human may spend thirty seconds reviewing a summary before clicking Approve.

That one click could release millions of dollars, terminate an employee, prescribe a medical treatment, deploy software into production, or trigger another chain of automated decisions. The physical effort required is almost nothing. The responsibility attached to it is enormous.

That’s the irony of AI.

The less work the human performs, the more responsibility the human may ultimately carry.

George Jetson pushed buttons because machines did the work. We’re beginning to push buttons because AI is doing the thinking. The joke from a 1960s cartoon is quietly becoming one of the defining realities of modern business.

AI may generate the recommendation, but the human authorizes reality.

The Rubber-Stamp Human

The great weakness in human-in-the-loop systems is that the human may remain in the process without providing meaningful human judgment.

Imagine an employee who reviews hundreds of AI recommendations each day. At first, the employee pays close attention. They inspect the reasoning, check the facts, and question unusual conclusions. But the system is usually correct, and the queue keeps growing.

Eventually, management begins measuring how quickly recommendations are processed. Rejecting the AI requires more explanation and creates more work. The employee becomes accustomed to seeing correct results, and the approvals become routine.

At that point, the employee is no longer seriously evaluating the AI. The employee is simply processing its decisions.

A human in the loop is not protection when the human has become a ceremonial rubber stamp.

This is made worse by automation bias, which is our tendency to trust automated recommendations, especially when the system appears more knowledgeable than we are. The more capable AI becomes, the harder it may be for a person to challenge it.

If the system has been correct 999 times, the reviewer may not notice the one dangerous recommendation hidden among them. Even when they do notice something unusual, they may not feel qualified to disagree.

This creates a troubling arrangement in which the AI has the knowledge, the company has the power, and the employee carries the blame.

The flip side of this is any kind of work. Anything that requires thinking, reasoning, and decision making, it will eventually be replaced by AI. So once the machine gets it 999 times right, the human is not needed to push Approve.

The Collapse of the Button Pusher

Of course, this arrangement won’t last forever.

Today, humans still approve AI because the systems occasionally make mistakes. But what happens when an AI is right 99.9% of the time? What happens when the human reviewer becomes the least reliable part of the process?

At some point, constantly asking for approval stops making sense.

The button pusher disappears.

This won’t happen everywhere overnight, and it certainly won’t happen in industries where human accountability is legally required. But across many businesses, the approval layers will begin collapsing.

First, the employee whose job is simply to approve routine work becomes unnecessary.

Then their supervisor, whose primary responsibility was reviewing those approvals.

Then another layer of management.

Eventually, organizations begin flattening because fewer people are needed to coordinate the work. AI isn’t just replacing individual jobs; it’s replacing the communication, coordination, and approval structures that companies have spent decades building.

Responsibility doesn’t disappear. It simply moves upward.

Each time a layer is removed, the responsibility attached to that layer moves to the next person above it.

Eventually, much of that responsibility lands with the entrepreneur or owner.

We’re already seeing early examples of this. Companies with remarkably small teams are generating revenue that would have required hundreds of employees only a few years ago. AI allows one person’s judgment to direct an extraordinary amount of production.

The future may not belong to the person who performs the most work. It may belong to the person trusted to make the best decisions and accept responsibility for them.

The Rise of the AI Follower

This leaves a massive new type of person emerging. Not the button pusher, but the AI follower.

The button pusher at least believed they were making a decision. They may have reviewed the work poorly or approved it too quickly, but they understand that the final choice is theirs.

The AI follower gives up that role. They ask the system what to do and then do it.

Which candidate should I hire? Which employee should I fire? Which investment should I make? Which customer should I reject? What should I write? What should I believe? How should I respond?

The system gives an answer, and the person follows it because the AI appears more informed, more objective, and more capable than they are.

This will be especially tempting because AI is built from the numbers, documents, decisions, patterns, and behavior that humans have produced. It can speak with the apparent authority of accumulated human knowledge. It may analyze more information than any person could read and identify patterns no person could easily see.

But it is important to understand what those patterns represent.

AI reflects what human beings have done, recorded, measured, rewarded, and preserved. It may be very good at identifying what usually happens. That does not mean it knows what should happen.

A system trained on past hiring decisions may repeat old prejudices. A system trained to maximize profit may recommend conduct that is effective but harmful. A system designed to increase engagement may promote anger or fear because those emotions produce strong results.

AI can turn human history into a recommendation. It cannot guarantee that human history deserves to be followed.

The New Valuable Employee

The most valuable worker in the AI era may not be the person who produces the most output. It may be the person who asks the right questions, recognizes weak assumptions, understands context, detects risk, knows when to escalate, and refuses recommendations that should not be followed.

The most dangerous employee may be the one who always agrees with the AI.

The most valuable may be the one who knows when not to.

That person will need enough knowledge to recognize when the system may be wrong, enough independence to resist its authority, and enough courage to accept the consequences of disagreement.

This will not be easy. Rejecting a weak recommendation from another person is one thing. Rejecting a system that has read more, analyzed more, and been correct more often than you have is something else entirely.

The future will require a new kind of discipline. We should use AI, learn from it, and respect its abilities, but we should never confuse intelligence with moral authority.

The machine may be better at producing the recommendation. The human must remain capable of refusing it.

The approval interface will probably become increasingly simple. A short summary, a green button, one click, and instant action.

Behind that button, however, may sit thousands of AI-generated decisions. It may control millions of dollars, a person’s employment, a patient’s treatment, a customer’s privacy, a public statement, a military action, or a permanent change to critical software.

The simpler the interface becomes, the easier it is to forget the complexity and consequence hidden behind it. Moral disengagement.

Good interface design usually removes friction, but some decisions should contain friction. A life-changing action should not feel like accepting a calendar invitation. A system capable of producing enormous consequences should not hide those consequences in the name of convenience.

The approval button should not erase moral seriousness. It should concentrate it.

The Jetsons may have been more perceptive than we remember. The joke was that George’s work had been reduced to pushing a button, but the show also understood that the person assigned to push it could become the person expected to answer for what happened next. In the age of AI, that joke is becoming less funny. The work behind the button may be performed by machines, while the human at the console is given the title, the authority, and possibly the blame.

The person at the console will not merely be operating a machine. They will be deciding what the machine is allowed to do. They will be expected to understand its power, resist it when necessary, and answer for the consequences.

AI may do the work, but responsibility begins when a human pushes the button.