Knowledge Work Versus Procedural Work
When people discuss AI and employment, the conversation usually revolves around one question: Which jobs will AI replace? While being a valid question, perhaps it is not the most important one. The first major wave of white-collar job losses may not require artificial intelligence capable of replacing accountants, analysts, bankers, lawyers or managers outright. A much simpler mechanism may be sufficient. AI is making traditional (dumb) software dramatically cheaper and faster to create. And that means many workers may not lose their jobs to AI at all. They may lose their jobs to ordinary software that AI made cheap and easy enough to build.
For decades, countless business processes remained manual not because they were inherently difficult to automate, but because automating them with traditional software was not economically worthwhile. A company might employ six people to process forms, reconcile spreadsheets, transfer information between systems, produce reports, check documents and send routine communications. Management may have known for years that much of this could theoretically be automated. But a traditional software project might have required requirements gathering, analyst support, a development team, testing, project management and months of implementation and support.
If automating a process costs $300,000 and saves $100,000 per year, management has a decision to make. If the same process can eventually be automated for $20,000, or perhaps built internally by one capable employee assisted by AI, then the calculation becomes redundant. The automation threshold collapses. And when that happens, an enormous amount of work that was previously protected by the economics of software development suddenly becomes exposed.
This is reminiscent of that technically inclined employee who once wrote Excel macros to automate repetitive tasks. Except now, with AI, that same employee can potentially build surprisingly capable software at a fraction of the time and cost it would have taken in the old days.
How White Is Your White-Collar Work?
We tend to classify jobs according to where they are performed rather than what the worker actually does. Someone sitting behind a computer in an office, holding a university degree and working in Excel, SAP or some proprietary corporate system is automatically classified as a white-collar worker. But much of what happens inside modern offices is not really knowledge work. Instead, it is procedural work.
- Compare two values.
- Update a field.
- Forward a document.
- Escalate if a threshold is exceeded.
- Repeat.
The environment may look white collar, but the underlying structure sometimes resembles an assembly line. The worker is not primarily being paid to decide what should happen, but to execute a process that somebody else designed. The distinction between knowledge work and procedural work may become far more important than the old distinction between white-collar and blue-collar employment. And AI changes the economics of both knowledge and procedural work in several ways at once. It can help analyze workflows, generate code and documentation, write tests, and modify existing applications. The significance is not merely that programming becomes faster. The entire life-cycle of creating small pieces of business software becomes cheaper and much faster.
The Secondary Effect of AI
Most discussions about AI displacement focus on the primary effect: AI learns to perform a task previously performed by a human. But the secondary effect may prove equally important.
Imagine a company where five employees spend much of their day performing a repetitive internal process. The company does not need an AI model to perform that process forever. It may only need AI temporarily to help someone understand the workflow and write deterministic (traditional, dumb) software that performs it. The final system might contain very little artificial intelligence. It might effectively say:
- If this happens, retrieve that information.
- If the amount exceeds this value, request approval.
- Check if these fields match, then continue.
- If they do not match, flag the transaction.
- Update the database.
There is nothing particularly intelligent about such a system; it is just ordinary, dumb software. The difference is that software that previously required an expensive development project may increasingly be created in weeks if not days. This creates an uncomfortable possibility for some workers: Their competitor is not some future superintelligence. Their competitor is an if statement. AI merely made the if statement cheap and easy enough to write.
The Workplace Was Never as Equal as the Org Chart Suggested
There is another aspect of this transition worth discussing. Traditional office work has often been surprisingly poor at distinguishing between very different levels of capability: Two employees may have the same title, the same grade, the same formal responsibilities, perhaps even similar salaries. Yet the way they relate to their work may be completely different.
Both may perform exactly the same visible tasks. One employee understands very little beyond the instructions required to complete them. But the other understands why the process exists, recognizes unusual cases, questions rules that no longer make sense, thinks about consequences, notices inefficiencies, can explain the process to somebody else, and even can redesign parts of it.
Both appear productive because both produce the expected output. The traditional workplace frequently treats them as interchangeable. In some ways, that has always been unfair.
Organizations are good at measuring attendance, titles, tenure, transactions processed, tickets closed and forms completed. They are often much worse at measuring understanding. The employee who thinks deeply about a process may receive little additional recognition for doing so. The employee who takes initiative may even create more work for themselves. The person who notices a broken process becomes the person asked to fix it. The person who understands the system becomes the person everybody calls when something fails.
Meanwhile, an employee who carefully avoids additional responsibility may remain within the same organizational category. For years, bureaucracy has been remarkably good at hiding these differences. AI may begin making them more difficult to hide.
Three Characteristics of Vulnerable Work
The roles most vulnerable to the secondary effects of AI may share three characteristics.
1. Procedural Work
This is the straightforward one: The more easily a job can be expressed as a repeatable sequence of decisions and actions, the easier it becomes to automate.
- “If X happens, check Y.”
- “If Y meets condition Z, do A.”
- “Otherwise do B.”
This does not mean every procedural role disappears. Some apparently repetitive processes contain difficult exceptions, regulatory obligations or situations requiring genuine judgment. But the more straightforward the workflow, the more likely it will be to replace human execution with software automation.
2. Low Organizational Leverage
The second characteristic is more subtle. Some employees execute processes while others understand them, and maybe can even improve them. Some know why the process exists in the first place. These people may have the same job title, work in the same department and appear almost identical on an organizational chart. But their organizational leverage is very different.
A worker who processes 100 transactions per day creates a certain amount of output. A worker who redesigns the process so that software handles 90% of all transactions changes the economics of the department. This distinction becomes increasingly important as AI amplifies what capable individuals can accomplish.
Here’s an alternative take for the same situation: A department that once required ten people may eventually require three. Three employees who properly understand the job, can handle exception cases and are equipped with better software can now produce the output of ten.
The fundamental employment question therefore becomes less: Can AI perform my job? and more: Which people will my organization need to produce the same output, after software automation kicks in?
3. Low Adaptability
The third characteristic is uncomfortable to discuss, but difficult to ignore.
Large organizations have always contained employees whose primary career strategy is avoiding exposure.
- Do not volunteer.
- Do not take ownership.
- Do not become responsible for difficult problems.
- Learn only what is necessary.
- Follow the procedure.
- Remain unnoticed.
- Reach the end of the day.
This is not always laziness. Sometimes organizations themselves train people into this behavior by punishing initiative, rewarding conformity or giving little benefit to those who take responsibility. But regardless of how the behavior develops, it becomes increasingly risky in an environment where workflows can be redesigned quickly.
When processes are changing, the employee who merely knows how to follow yesterday’s procedure becomes less valuable than the employee who can learn and adapt to tomorrow’s process. This is not simply a question of intelligence, but a question of adaptability.
Can someone learn a new system, use unfamiliar tools, take responsibility for outcomes rather than tasks, identify what should be automated instead of fearing automation?
The worker who refuses to change may become vulnerable even if AI never becomes capable of performing that worker’s entire job. AI only needs to change the surrounding economics enough for the organization to stop needing the role in its existing form.
The Executor, the Operator and the Owner
One way of understanding this transformation is to divide workers into three broad categories. These are not job titles; they are relationships to work.
The Executor knows what to do. The Operator knows how it works. The Owner knows why it exists.
The Executor
The Executor asks: “What should I do?”
The Executor follows an established process. The value lies primarily in performing the steps correctly and consistently. This is the type of work most directly threatened when procedural automation becomes cheaper. If the process can be documented, formalized and encoded, the need for repeated human execution declines.
The Operator
The Operator asks: “How does this process actually work?”
Operators understand the process beyond its formal description. They know what normally happens, what occasionally happens and what happens when things go wrong. They understand the exceptions, the ins and outs of the process, such as:
- which customer always submits the wrong document,
- that one supplier formats its invoices differently,
- what seemingly harmless error eventually causes a serious problem three steps later,
- who must be called when the documented process stops making sense.
Operators possess something organizations frequently underestimate: tacit knowledge.
The Owner
The Owner asks: “What are we actually trying to accomplish?”
The Owner understands the objective rather than merely the process.
- Why does this workflow exist?
- Which steps provide value vs which exist because of historical accident?
- What can be removed or automated?
- Which exceptions genuinely require human judgment?
- What should happen when the system encounters something nobody has experienced before?
AI may make Owners dramatically more powerful. An employee who understands a business process deeply and can use AI to redesign, automate and monitor it may suddenly have enormous organizational leverage. This creates the possibility of a growing divide inside supposedly identical professions.
Two people may both carry the title “Senior Operations Analyst.” One executes a 37-step procedure. The other understands why the procedure exists and can redesign it. Their job titles have been the same. Their future economic value may be very different.
AI May Expose a Difference the Old Workplace Concealed
This is where AI may make the old difference economically visible. Traditional organizations could treat two employees as roughly interchangeable because both produced similar visible outputs. AI introduces another variable: leverage. Once capable employees can turn understanding into automation, analysis and software, the difference between following a process and improving it begins to affect what a single person can accomplish.
The employee who understands the process can now do something with that understanding. They may be able to improve a workflow, build a small internal application, automate repetitive steps, connect systems, create a management dashboard, and analyze a decade-old backlog of records in a single week.
What previously required access to a development department, a budget and management approval may increasingly be within reach of an ordinary knowledge worker. This changes the balance. AI can give the thoughtful employee something that traditional bureaucracy often denied them: the ability to turn understanding into execution.
In that limited sense, AI may democratize organizational leverage. It does not automatically make organizations fair. It does not guarantee that the best people will be recognized. And badly managed companies may even use AI primarily as a crude cost-cutting instrument. But AI will weaken one old organizational advantage: the ability to look indistinguishable from the person next to you while contributing much less understanding, initiative and ownership.
Before AI, two employees might produce the same visible spreadsheet. After AI, one of them may redesign the process that produced the spreadsheet. That difference becomes much harder to ignore.
Execution Gets Cheaper
This leads to a broader shift. For much of the modern corporate era, organizations paid people to execute processes because execution was expensive.
- Information had to be gathered manually.
- Systems had to be updated manually.
- Reports had to be generated manually.
- Coordination required meetings, emails and administrative work.
AI reduces the cost of many of these activities. And as execution becomes cheaper, judgment, ownership and problem definition become proportionally more valuable.
- Knowing how to perform a task matters less when a machine can perform the task.
- Knowing which task should be performed matters more.
- Recognizing that the wrong problem is being solved matters more.
- Understanding the consequences of an automated decision matters more.
- Knowing when the system has encountered something outside its assumptions matters more.
- And taking responsibility for the result matters more.
The paradox is that AI may increase the relative value of distinctly human judgment at precisely the same time that it reduces the amount of human labor required.
The Corporate Memory Trap
There is, however, a serious danger for organizations that move too aggressively. Companies looking at a department may see ten employees performing what appears to be largely repetitive work. Management may reasonably conclude that new software can reduce the requirement to three. But the seven employees leaving the organization do not merely take labor with them: they also take tacit knowledge with them. Corporate memory loses part of that invisible layer beneath the documented workflow.
Officially, the process says A, B, C and D. In reality, A usually works. B works unless the transaction comes from a particular customer. C requires someone to remember a regulatory exception introduced four years ago. D sometimes fails because two systems interpret a field differently. And when an unusual case appears, everyone asks that one experienced person, because she has been doing this for fourteen years and somehow can lead others in every unexpected case. From the perspective of the actual organization, that person may be part of the corporate operating system.
If the company automates the documented process and removes the people who understand the undocumented one, it may discover the missing knowledge only after those people have left. This creates an important distinction between removing labor and removing capability. An aggressive automation program can therefore become self-defeating.
The company successfully removes the people performing routine work, only to discover later that those same people were handling edge cases, maintaining informal controls, compensating for broken systems and preserving institutional memory. In a bad implementation of AI, the company may aim to become more efficient with AI, while simultaneously making itself less knowledgeable.
Automate the Process, Not the Knowledge
The sensible objective therefore should not simply be headcount reduction. It should be organizational learning. Before automating a process, companies need to understand what their experienced employees actually know.
- Where are the exceptions?
- Which decisions depend on judgment?
- Which controls exist only because experienced employees remember to perform them?
- Which parts of the workflow are historical baggage?
- Which parts protect the organization from failures nobody notices because those failures are currently being prevented?
AI may actually help with this process. Experienced employees could work alongside AI systems to document workflows, identify exceptions, model decision paths and preserve institutional knowledge before automation occurs. That produces a much healthier transition. Routine execution becomes automated while tacit knowledge becomes explicit.
Operators can become designers, supervisors or exception handlers. And organizations avoid firing the last person who knows why the strange-looking step in the middle of the workflow exists.
Who Gets Slashed?
While the conventional answer would be that AI threatens white-collar workers, the term “white-collar worker” may be far too broad a category to be useful. The more meaningful distinction may be between procedural work and knowledge work.
- Executing a process vs understanding it.
- Following instructions vs defining outcomes.
- Operating inside a system vs being capable of redesigning the system itself.
The workers most exposed may not necessarily be those in a particular profession. They may be those whose contribution is primarily procedural, whose organizational leverage is low and whose willingness to adapt is limited. Likewise, the workers who benefit most may not necessarily be AI specialists or technologists. They may simply be people who understand their work deeply enough to ask better questions about it—the kind of questions that separate the Executor from the Operator, and the Operator from the Owner.
I also see a fairness in the change ahead. For years, the workplace could place the person who merely executed the procedure beside the person who understood, questioned, improved and owned it, and could call them equivalent. AI may not eliminate that unfairness, but it will likely make the difference much harder to conceal.
AI does not need to become an artificial employee sitting at every desk to transform the workplace. It only needs to make automation cheap enough that companies begin asking a question they previously could not afford to ask:
Why is a human still doing this?
And once that question is asked, another follows immediately:
Who actually understands why we are doing it at all?