When Everything Needs Approval, Nothing Really Improves

Jul 20, 2026

10 min read

1 views

89%

The hidden cost of control: speed dies before trust does

What do a contracting bottleneck in homeland security and a wave of AI productivity hype have in common? More than it first appears. Both point to the same uncomfortable truth: modern institutions often confuse visibility with value, and control with competence.

On paper, tighter oversight sounds responsible. If a powerful agency can sign off on fewer contracts without leadership approval, surely that prevents waste. If AI can generate text, moderate content, and automate routine work, surely productivity should rise. But the deeper pattern is more unsettling. In both cases, organizations keep reaching for tools that promise leverage while building systems that make real work harder to do, less visible, or more fragmented.

That is the paradox at the center of our age of automation and administration: we keep trying to create efficiency by inserting more gates, more screens, more layers, and more invisible labor. The result is not always cheaper operations. Sometimes it is a slower institution, a more exhausted workforce, and a system that looks smarter from the outside while becoming more brittle inside.

The modern organization is often optimized to look controllable, not to become more capable.

The bureaucracy trap: when oversight becomes the bottleneck

There is a seductive logic behind tighter sign off rules. If a leader approves more decisions, the organization becomes more disciplined. Exceptions get caught. Spending gets scrutinized. Power becomes centralized enough to prevent drift. In theory, this should improve governance.

But there is a threshold beyond which oversight changes from a safeguard into a choke point. Once routine decisions need escalation, the organization stops behaving like a distributed system and starts behaving like a single crowded desk. Contracts wait. Vendors stall. Teams hesitate. People begin to optimize not for mission outcomes, but for what will survive the next layer of review.

This is the first mental model worth keeping: every approval gate has a shadow cost. The obvious cost is time. The deeper cost is initiative. If a manager needs permission for every meaningful move, then local expertise becomes ornamental. People who know the work best end up spending their days anticipating the preferences of people who do not.

That dynamic is not unique to government. It appears in large corporations, hospitals, universities, and even startups once they cross a certain complexity threshold. A company can become so afraid of bad decisions that it invents a machine for preventing decisions. The machine may be rational at the level of risk control, but irrational at the level of throughput.

A useful analogy is air traffic. You want centralized coordination in the sky, because collisions are catastrophic. But you do not want every plane waiting for a personal blessing from headquarters before changing altitude by a thousand feet. The system works because rules are delegated, not because every move is manually approved.

That is the real question every institution must answer: Which decisions require hierarchy, and which merely require standards? The failure mode of modern bureaucracy is treating both as if they are the same thing.

AI’s dirty secret: the invisible workforce never disappeared

Automation sells itself with a clean story. A machine takes over the repetitive task. A chatbot drafts the content. A model sorts the requests. A robot handles the mess. Humans are liberated to do higher value work, and productivity rises.

Yet the actual structure of automation is messier. Instead of removing labor, it often relocates labor into less visible places. Someone still labels the data. Someone still reviews the edge cases. Someone still cleans up the outputs. Someone still catches what the model misses. In the most ambitious systems, workers do not vanish. They are merely moved behind the curtain, where the user can no longer see them.

That matters because visibility shapes how labor is valued. When customers see a smiling interface, they imagine machine autonomy. When executives see lower headcount in one visible department, they imagine efficiency. But the work may simply have been transferred to contractors, moderators, annotators, or support staff who are more precarious and more replaceable.

This is the second mental model: automation often fragments work instead of eliminating it. It breaks a job into visible output and invisible maintenance. The visible output gets celebrated. The invisible maintenance gets discounted. Yet the maintenance is where quality lives.

Think of a restaurant. The dining room is the theater. The kitchen is the machine. If a company decided to “automate” the restaurant by hiding the kitchen farther from view, diners might feel the experience is more seamless. But if the cooks, dishwashers, and prep staff are still doing the labor, only now under tighter pressure and lower recognition, the operation is not magically efficient. It is just more opaque.

The same thing happens with AI systems that appear to write, filter, and decide on their own. Underneath them are human raters, safety teams, moderators, and reviewers confronting the unpleasant edge cases that machines are not built to absorb. The less visible those people are, the easier it becomes to pretend the system is purely technological.

But opacity has a price. When labor disappears from view, it also disappears from management attention. That is how companies can celebrate automation while quietly underinvesting in the very human expertise that keeps the automation from failing.

Why productivity keeps disappointing: we built systems that scatter effort

For years, a common promise followed every new wave of software. Digitize the workflow, and productivity will jump. Automate the back office, and output per worker will rise. Connect everything, and coordination will get easier.

Yet productivity growth has remained stubbornly weak. That is not just a statistical curiosity. It is evidence that many digital systems are not reducing work so much as redistributing friction.

Here is how that happens. A new tool speeds up one step in a process, but creates three more review steps. A chatbot drafts customer replies, but humans now spend more time checking for errors and tone. A centralized contract approval rule reduces bad spending, but creates delays so long that teams improvise workarounds. The burden is never destroyed. It migrates.

This is why the classic question, “Does the tool save time?” is too shallow. The better question is: Where does the time go? If a tool saves 20 minutes for one person but forces ten other people to spend two minutes each validating, routing, correcting, or reentering the same information, the organization may end up worse off.

A practical framework helps here: measure technology not by output alone, but by coordination tax. The coordination tax includes all the time spent on approvals, exceptions, handoffs, audits, review loops, and labor that exists purely because the system created uncertainty or opacity. A tool that lowers direct labor but raises coordination tax may still feel modern while lowering total capability.

This is one reason many organizations experience what looks like digital progress but feels like operational fatigue. Employees are not merely doing their core work. They are also acting as translators between software systems, compliance demands, and managerial anxieties. The result is a workplace where people are busier but not necessarily more effective.

Productivity does not fail only when machines are weak. It also fails when institutions make humans spend their time compensating for machines.

The deeper connection: institutions are becoming anti-fragile in appearance and fragile in practice

At first glance, stricter oversight and AI automation seem like opposite trends. One adds human approval. The other removes human visibility. But they are actually cousins. Both are attempts to manage complexity by abstracting away the messy middle.

In the oversight case, leadership abstracts the organization into permission levels. The messy middle of judgment gets pulled upward. In the automation case, companies abstract labor into outputs. The messy middle of human effort gets pushed downward and hidden. In both situations, the institution becomes easier to narrate but harder to operate.

That is the structural danger. Systems become more legible to the top and less workable in the middle. Leaders believe they are reducing uncertainty because they can now see more of the pipeline, the dashboard, or the approval tree. But what they actually reduce is the capacity of people closest to the work to respond quickly and intelligently.

This creates a paradox of modern management: the more we centralize control or automate interfaces, the more we risk losing the tacit knowledge that makes the system functional. Tacit knowledge is the kind you cannot fully write down. It lives in seasoned judgment, informal shortcuts, and the ability to notice what is unusual before it becomes a crisis.

When a contracting office forces tiny deals through the top, it signals distrust in lower-level judgment. When an AI stack hides content moderation or data labeling behind a glossy interface, it signals that labor is mainly a cost to be concealed. In both cases, institutions teach themselves the wrong lesson: that human discretion is dangerous when in fact it is often the only thing preventing rigidity from turning into collapse.

The danger is not merely inefficiency. It is organizational atrophy. A team that never gets to decide stops learning how to decide. A workforce that is never recognized for invisible labor stops building pride in craftsmanship. A system that only sees the output and not the process gradually loses the ability to improve either.

A better operating principle: make the invisible legible, not the visible more controlled

If the problem is not too little oversight or too much AI, what should leaders do instead?

The answer is not to abandon control. It is to redesign where control lives. The goal should be to make the invisible parts of work legible, not to impose new gates on every visible transaction.

That means two things.

First, push authority closer to expertise. If a recurring decision can be governed by standards, templates, or thresholds, do that. Save executive approval for true exceptions. In other words, centralize the unusual, delegate the routine. This keeps the organization responsive without abandoning accountability.

Second, surface the hidden labor behind automation. If AI is involved in the workflow, track the human maintenance required to keep it honest, safe, and useful. Count the hours spent reviewing outputs, fixing errors, handling edge cases, and retraining systems. If those costs are rising, the tool may be shifting labor, not reducing it.

A healthy institution should ask three questions about any new layer of control or automation:

Does it reduce the number of decisions, or only relocate them?
Does it reduce total effort, or only hide effort from view?
Does it expand the ability of people closest to the problem, or narrow it?

These questions are uncomfortable because they puncture the illusion that process improvements are always improvements. Sometimes the best indicator of progress is not the sleekness of the interface, but the amount of judgment the system still trusts its people to exercise.

Key Takeaways

Measure coordination tax, not just speed. A faster workflow that creates more approvals, reviews, and exception handling may be a net loss.
Treat invisible labor as core labor. If humans are still cleaning, checking, moderating, or correcting an AI system, that work is not peripheral. It is the system.
Delegate routine decisions and centralize only exceptions. Oversight should protect standards, not absorb every ordinary transaction.
Watch for productivity theater. A dashboard, chatbot, or approval rule may look impressive while merely moving effort around.
Preserve local judgment. Teams improve when people closest to the work are trusted to decide within clear boundaries.

The real question is not whether humans are replaced

The most important myth in both bureaucracy and automation is that progress means removing people from the loop. In reality, the loop always remains. The question is whether the people inside it are empowered experts, invisible laborers, or exhausted approvers.

That reframes the debate entirely. The challenge is not to build systems with fewer humans. It is to build systems where human intelligence is not wasted on needless permissioning or hidden cleanup. In that sense, the future of efficient institutions is not fully automated and not fully centralized. It is properly visible.

The most advanced organization will not be the one that controls everything from the top, nor the one that hides all labor behind software. It will be the one that knows where judgment belongs, where labor belongs, and where friction is simply a sign that the system has forgotten what work is for.

When everything needs approval, nothing really improves. When everything is automated, nothing is truly gone. The deepest competitive advantage is not more control or more machine intelligence. It is the ability to design institutions that respect the full cost of making things happen.