When AI Learns Your Work, It Also Reveals What Work Is Really For

Jul 01, 2026

9 min read

2 views

88%

The Strange Question Hidden Inside AI at Work

What if the most important thing generative AI is measuring is not intelligence, but agency?

That sounds like a philosophical detour, but it points to a practical shock that many knowledge workers are already feeling. The tasks people most often ask AI to help with are the very activities that once made their work feel distinctly human: gathering information, writing, explaining, advising, and teaching. In other words, AI is not just automating labor. It is moving into the territory where many people experience themselves as agents, the ones who form intentions, choose means, and shape outcomes.

That creates a deeper tension than simple job replacement. If agency is the feeling and function of being the one who initiates action toward a goal, then AI is not merely changing how we work. It is changing where the action seems to come from.

The real disruption is not that machines can do tasks. It is that they can now occupy the parts of work through which people most clearly experience themselves as authors of outcomes.

This is why the conversation about AI feels strangely personal even when the metrics are occupational. We are not only asking which jobs are at risk. We are asking which parts of work still belong to the worker as a decision maker, not just as a labor performer.

Agency Is Not a Feeling, It Is a Structure

Most people think of agency as a vibe, a sense of ownership over what they do. But agency has a stricter shape. It involves intentionality, the ability to represent a future goal; equifinal variability, the ability to reach that goal through different routes; and rationality, choosing the most efficient means available in context.

That matters because it means agency is not merely “doing things.” A machine can produce output. A human agent, by contrast, selects among possibilities in light of a goal. The point is not just action, but goal-directed control under uncertainty.

This distinction helps explain why some AI use feels liberating while other uses feel unsettling. If a tool helps you draft an email faster, it may amplify your agency. If it begins deciding what the email should say, how it should be framed, and when it should be sent, it begins to absorb the structure of agency itself.

Think about a sales manager who uses AI to summarize customer calls. At first, the tool is a clerk, sorting raw material so the manager can act. Then it becomes a strategist, recommending the next best message. Then it becomes a coach, telling the manager how to persuade. At each step, the tool moves closer to the center of human agency, not just the edges of productivity.

The crucial insight is this: agency is not binary. It is distributed across tasks, tools, and environments. Work is not simply what gets done. Work is a system for preserving and directing agency.

Why Writing, Advising, and Information Work Are So Vulnerable

The occupations with the highest AI applicability are often not the most physical, but the most linguistic. Computer and mathematical work, office and administrative support, and sales all sit near the same fault line: they involve collecting information, translating it, and guiding decisions.

That should not surprise us. These jobs are built around a repeated sequence:

Perceive what is going on.
Interpret it.
Communicate it.
Recommend a next step.
Help execute the decision.

Generative AI is especially strong at steps 2 through 4. It can synthesize, explain, draft, and advise at scale. That means it does not merely save time. It compresses the distance between raw data and a plausible action.

A traditional analyst might spend an afternoon gathering information, comparing options, and writing a memo. AI can now produce a first pass of that memo in seconds. A teacher may spend hours preparing explanations, examples, and feedback. AI can generate those too. A sales representative may have to research prospects, craft outreach, and shape responses. AI can assist with all of it.

The deeper issue is not just efficiency. It is that the labor of becoming oriented to a situation is being automated. That labor used to be one of the main ways workers exercised judgment. When that work is externalized, the worker risks becoming a reviewer of machine-made possibilities rather than the originator of a course of action.

This is the hidden occupational implication of generative AI: it targets the scaffolding of agency before it targets the final act.

The Agency Spectrum: From Tool to Co-Author to Proxy

A useful way to think about AI in the workplace is not as a single category, but as a spectrum of agency transfer.

1. Tool mode

AI expands human agency without competing with it.

Example: a lawyer uses AI to search case law faster. The lawyer still chooses the question, weighs the relevance, and decides the argument.

2. Co-author mode

AI contributes substantive material that shapes the final output.

Example: a marketer uses AI to generate several campaign concepts. The marketer curates, edits, and combines them, but the shape of the message is now partially machine-suggested.

3. Proxy mode

AI begins making intermediate decisions on behalf of the user.

Example: a customer service system triages inquiries, drafts responses, and recommends resolutions before a human ever sees the case.

4. Substitution mode

AI effectively performs the activity with only minimal human oversight.

Example: routine report writing, basic scheduling, standard proposal generation, or templated advisory work.

The danger is not that all work will rapidly jump to substitution mode. The more subtle risk is that many jobs will drift there piecemeal. The worker remains formally in charge, but their role becomes increasingly ceremonial, approving outputs that were largely determined elsewhere.

That is where the psychology of agency becomes essential. People do not only need outcomes. They need the experience that their actions are means to an end they genuinely selected. Without that, work can become efficient but alienating, productive but hollow.

A job can survive automation and still lose its soul if it stops giving people meaningful control over what happens next.

The Paradox: AI Can Increase Productivity While Shrinking Purpose

The obvious promise of AI is that it frees people from drudgery. The less obvious possibility is that it also frees people from the very process through which they experience competence and authorship.

This is the paradox of modern work: many of the most annoying tasks are also the tasks that make us feel useful. Researching a problem, drafting a response, checking assumptions, revising a pitch, explaining a concept to a colleague, these are not just chores. They are enactments of agency.

When AI takes over those steps, a worker may feel more efficient but less necessary. The output remains, but the inner movie changes. Instead of “I figured this out,” the feeling becomes “I approved something that was figured out for me.” That distinction matters because motivation follows perceived authorship.

Consider a junior consultant. Before AI, they might spend hours assembling slides, analyzing data, and writing recommendations. The process is tedious, but it teaches them how to think. After AI, they can produce a polished deck quickly. Yet if they never wrestle with the material, they may learn how to format expertise without building expertise itself.

The same risk exists in education. A student who uses AI to produce essays may submit stronger prose, but if the tool also removes the need to form a thesis, choose evidence, and refine logic, the student loses the apprenticeship of thinking. The output looks better while the internal machinery weakens.

This is why the most important question about AI is not, “Can it do the work?” It is, “What kind of person does the work make, and what kind of agency does it preserve?”

What Good Organizations Will Protect

If agency is central to work, then the smartest organizations will not simply automate more. They will design for agency preservation.

That means distinguishing between tasks that should be accelerated and tasks that should remain human because they develop judgment. It also means recognizing that not all efficiency is value creating. Sometimes friction is not waste. Sometimes it is the price of learning, ownership, and discernment.

A strong organization should ask three questions for every AI deployment:

Does this tool remove mechanical effort or meaningful judgment?
Does it support better decisions, or just faster outputs?
Does it deepen a worker’s competence, or merely hide the work from them?

Imagine two customer support teams. In the first, AI drafts replies, routes cases, and suggests policies, while humans review exceptions and learn from the patterns. In the second, AI handles almost everything, and humans only intervene when the system fails. The first team becomes more capable over time. The second team becomes a human exception layer, valuable only when automation breaks.

The difference is not technical. It is organizational philosophy. One design treats AI as an amplifier of human agency. The other treats it as a replacement for human participation.

Companies that understand this will preserve the parts of work that create judgment, not because they are nostalgic, but because those parts are the source of long-term adaptability. A workforce that no longer knows how to reason, write, explain, and choose is not just less fulfilled. It is less resilient.

Key Takeaways

Do not ask only whether AI can do a task. Ask whether that task is where people exercise judgment, learn competence, and feel authorship.
Map your work by agency level. Separate tasks into those that should be automated, assisted, reviewed, or kept human.
Protect the apprenticeship tasks. The hardest parts of writing, analysis, and advising often build the intuition that makes experts valuable.
Use AI to compress labor, not responsibility. Let it speed up preparation, but keep humans responsible for framing the problem and choosing the direction.
Measure organizational health by retained agency. If people become faster but less capable of independent judgment, the system is degrading even if output rises.

The Real Future of Work Is a Fight Over Who Gets to Decide

The most revealing thing about generative AI is that it shows how much of white collar work is actually a choreography of agency. The value is not just in producing text, slides, or answers. The value is in the act of deciding what matters, which facts to trust, what story to tell, and what action follows.

AI is powerful precisely because it can enter that choreography so naturally. It can write, advise, summarize, and teach. But when it does, it should make us more alert, not less. It is forcing an old question into sharper focus: what parts of our work are we willing to delegate, and what parts must remain the seat of human intention?

The future will not belong to the people who simply use AI the most. It will belong to the people and organizations that understand a harder truth: productivity without agency is a brittle achievement. The highest form of work is not merely getting things done. It is remaining the kind of person, and building the kind of system, that still knows why they were worth doing in the first place.