AI and the Bullshit Job Problem: What Happens When the Economy Can No Longer Afford Fake Work

The Audit Has Started

In 2018, anthropologist David Graeber published *Bullshit Jobs: A Theory*. His argument was simple and uncomfortable: a large and growing share of modern employment consists of jobs that serve no meaningful economic function. The people doing them often know it. Many told him so, anonymously, in the research that preceded the book.

Graeber identified five archetypes: flunkies (who exist to make their boss look important), goons (aggressive roles whose primary function is to counteract other goons), duct tapers (who fix problems that should not exist), box tickers (who create the appearance of compliance without the substance), and taskmasters (who manage people who do not need managing).

He estimated that somewhere between a third and half of all jobs in developed economies fell into one of these categories.

The reaction was predictable. Economists pushed back. Business leaders dismissed it. LinkedIn was outraged. Then everyone went back to scheduling meetings about meetings.

Six years later, AI walked into the building. And it turns out to be a very efficient auditor.

What AI Actually Threatens

The conventional narrative about AI and jobs focuses on factory workers, truck drivers, and call centers. The data increasingly tells a different story.

The jobs most exposed to large language model displacement are not physical. They are:

• Information processing roles where the primary output is reading one document and producing another
• Coordination roles where the primary value is moving information between parties who could communicate directly
• Compliance roles where the primary function is checking boxes against rules that could be encoded in software
• Sales roles where the primary technique is information asymmetry — knowing things the buyer does not

Physical trades — plumbing, electrical work, HVAC, construction, nursing — are largely not on this list. You cannot prompt-engineer a leaking pipe. The hands-on economy is more durable than the paper-shuffling economy, and that is an uncomfortable inversion for a workforce that spent decades credentialing itself out of manual work.

The Legacy System Guardian

Every large organization has them. They are the person who truly understands the enterprise software system that the company bought in 2003 and cannot replace because too much institutional knowledge lives inside it. They have attended every conference the vendor runs. They speak its proprietary query language. They know which buttons break things.

The system exists not because it is good but because replacing it is politically difficult. The guardian's job exists because the system exists. Neither is truly justified by economic value — they are artifacts of organizational inertia and historical brand equity that no one has had the courage to unwind.

AI cannot immediately eliminate this role. But it can do three things that make it increasingly precarious:

First, AI-assisted migration tools are dramatically reducing the cost of modernizing legacy systems. The technical argument for keeping the old system is weakening every year.

Second, AI can now document and extract institutional knowledge from these systems at a speed that used to require months of consulting engagements. The knowledge moat that protected the guardian's role is being drained.

Third, new AI-native platforms are being built from the ground up without the technical debt that made the old systems irreplaceable. The organizations that adopt them will not have a legacy guardian problem.

The guardian has two exits: become the person who leads the migration, or wait until the migration happens without them.

The Sleazy Sales Rep

Not all sales is sleazy. Complex B2B sales involving genuine technical expertise, real relationship capital, and honest value creation is not going anywhere. AI cannot build the kind of trust that closes a seven-figure enterprise contract.

But a specific flavor of sales is genuinely under threat: the kind that depends on information asymmetry, manufactured urgency, and the buyer's inability to comparison-shop effectively.

Car dealership finance offices. Certain insurance products. Enterprise software sold into organizations where the buyer cannot easily evaluate alternatives. Managed services contracts with auto-renewal clauses buried in paragraph seventeen. Commission structures that reward extracting maximum value from a customer rather than delivering it.

These models worked because buyers were at an information disadvantage. AI eliminates that disadvantage. A buyer who can run a detailed competitive analysis in twenty minutes, understand contract terms without a lawyer, and identify the actual market rate for a service in seconds is not the same buyer who used to walk into a dealership and say "what's the monthly payment?"

The sales rep who survives this is the one who was never dependent on information asymmetry in the first place. The one who actually knows the customer's business better than the customer does. The one whose value is insight and advocacy, not friction and pressure.

The one who was gaming the information gap is going to find that gap closed.

The Professional Box Checker

This is the largest category, and the most politically sensitive.

Large organizations have created enormous compliance, quality assurance, audit, and governance functions over the past several decades. Some of this work is genuinely valuable: real risks managed, real fraud prevented, real standards maintained. But a significant portion of it is performative — elaborate rituals designed to demonstrate that due diligence was performed, regardless of whether it was.

The ISO certification that requires thousands of hours of documentation for a company of twelve people. The quarterly security training that everyone clicks through in four minutes. The vendor due diligence questionnaire that is forty pages long and asks questions the vendor answers with copy-paste boilerplate. The change management process that takes three weeks to approve a one-line configuration change.

These processes were designed by committees optimizing for defensibility over effectiveness. AI can perform many of the underlying checks faster, more consistently, and more accurately than the humans who are currently running them — and it can do it continuously rather than quarterly.

The box checker whose primary value is generating paper will find that paper getting much cheaper to generate. The one who actually understands what the boxes are supposed to be measuring — and whether the measurements are actually catching real problems — has a much more durable career.

The Blue Collar Reality Check

Here is the part that makes white-collar professionals most uncomfortable: the trades are doing fine.

Plumbers, electricians, HVAC technicians, welders, carpenters, medical technicians, and industrial equipment operators are not facing an AI displacement crisis. Their work requires physical dexterity in unstructured environments, real-time problem-solving in conditions that cannot be predicted in advance, and direct human interaction in situations that require trust and presence.

Median wages for skilled trades have been rising. The backlog for licensed contractors in most US cities is measured in weeks or months. Community college enrollment in trade programs is at record highs as a new generation reappraises the ROI of a four-year degree versus a two-year certification.

The cultural positioning of blue-collar work as lesser than white-collar work was always partly a fiction maintained by the class that benefited from it. AI is accelerating the correction of that fiction. The person who fixes your HVAC system when it fails in July has rarer and more immediately valuable skills than many professionals who have spent decades in offices.

This is not an argument that everyone should become a plumber. It is an argument that the labor market is revaluing skills more honestly than social status has historically allowed, and that the people who dismissed trades as inferior to desk work are going to find the market disagrees with them.

What Actually Survives

The honest answer is that judgment, relationships, and embodied skill survive. Everything else is on a depreciation schedule.

Judgment means the ability to make decisions in ambiguous situations where the rules do not apply cleanly. A contract lawyer who can see the novel risk that the standard template misses. A nurse who can tell that the patient's numbers look fine but something is wrong. An engineer who can read a codebase and understand not just what it does but what it was trying to do. These are not things you can automate, because they require a model of the world that is not captured in training data.

Relationships means genuine trust earned over time through consistent demonstration of competence and care. The salesperson whose customer calls them when something goes wrong, not when they want to buy something. The manager whose team follows them to the next job. The consultant whose client actually implements their recommendations. These cannot be automated because they are fundamentally about human beings choosing to rely on specific other human beings.

Embodied skill means the ability to do physical work in the real world. The demand for skilled trades is rising, not falling, as populations age and physical infrastructure requires maintenance. These jobs pay well, cannot be offshored, and are directly immune to the displacement hitting white-collar work. The cultural stigma attached to them is a lag indicator of a world that no longer exists.

The Uncomfortable Implication

Graeber's theory was always controversial because it implied that a significant portion of what the professional class does is not actually necessary. Most professionals rejected this, understandably, because it is not a comfortable thing to believe about your own work.

AI is running the empirical test. If a language model can do your job in twenty seconds, the question of whether it was a real job has been answered.

The response to this should not be despair. It should be honesty — about what parts of your role actually matter, what parts were always performative, and where real value creation lives in your field.

The professionals who survive and thrive in the AI era will be the ones who asked that question before the AI did. Who stripped out the self-justifying busywork and invested in the parts of their role that required genuine expertise, judgment, and human connection. Who moved toward the work they probably always knew was the real work.

That transition is uncomfortable. But it is also, arguably, overdue.

What To Do If You Are In One of These Categories

If you are a legacy system guardian: Start leading the migration. Document what you know in a format that makes you the architect of the new system, not the caretaker of the old one. Your institutional knowledge is valuable — but it needs to be portable.

If you are in information-asymmetry sales: Audit your value proposition honestly. Are you solving a genuine problem for your customers, or are you exploiting a friction point that AI is about to eliminate? Shift toward technical depth, customer outcomes, and relationships that would survive a fully transparent market.

If you are a professional box checker: Understand what the boxes were supposed to catch and whether they are catching it. The compliance professional who understands why the regulation exists, not just how to document compliance with it, is far more durable than the one who can fill out the form efficiently.

If you are none of these but work alongside them: The organizations that identify and eliminate their own bullshit jobs proactively will have significant structural advantages over the ones that wait for the audit to be forced on them. If you have standing to accelerate that conversation, now is the time.

The economy is not going to continue paying for work that does not produce value. AI has simply made the value calculation faster and more visible.

That is, depending on your perspective, a threat or an opportunity. Often both.