What Will AI Do to My Job?

Everyone is asking the same question. Will AI put me out of work? Will it make me more productive? Will it make me irrelevant? The answers you hear are either breathless optimism ("AI will create more jobs than it destroys!") or existential dread ("AI will replace 300 million jobs!"). Neither is helpful, because neither tells you anything about your job specifically.

There is a better way to answer this question, and it starts with a deceptively simple exercise: define what your job actually produces. Not your title. Not your responsibilities. Not your calendar. What measurable output does your work create in eight hours?

That question — what do you produce, and how? — is the production function. It is the oldest equation in economics, and it is the sharpest tool we have for understanding what AI will and won't do to your livelihood.

The Equation

One Person

Strip everything back to first principles. You are one person. You show up to work. You have some tools (a computer, a phone, some software, maybe a vehicle or a machine). You have your time (roughly eight productive hours). You have your skills (education, experience, judgement). And you produce something.

Your personal production function is:

Y = T × S × K

Where Y is your output, T is your time, S is your skill, and K is your capital (tools). This is deliberately simple. If you double your time, you double your output. If you get better tools, your output goes up. If you improve your skills, your output goes up.

A farmer in 1800 had T = 12 hours, S = knowledge of crops, K = a hand plough. His output: enough food to feed three to five people.^[1]

A farmer in 2024 has T = 8 hours, S = agricultural science, K = a GPS-guided combine harvester, precision irrigation, and satellite crop monitoring. His output: enough food to feed 170 people.^[2]

Same equation. Radically different K. The farmer didn't get smarter or work harder. He got better machines.

One Company

A company is a collection of individuals, each running their own production function, coordinated toward a shared output. The company's production function is:

Y_company = (Σ Y_individual) × A

Where A is the coordination multiplier — how well the company organises its people and capital. A well-run company gets more out of the same people and tools than a poorly-run one. A captures management quality, culture, processes, and organisational design.

Ford's assembly line in 1913 didn't change the individual worker's tools dramatically. It changed A — the way workers were organised. Production time dropped from 12 hours to 93 minutes.^[3] Same workers, same basic tools, radically better coordination.

A company can increase output by: hiring more people (more individual Y's), giving them better tools (increasing each person's K), improving their skills (increasing S), or improving coordination (increasing A).

One Country

An economy is a collection of companies, each running their own production function, operating within shared institutions. The country's production function is:

Y_country = (Σ Y_company) × I

Where I is the institutional multiplier — rule of law, infrastructure, education systems, regulatory frameworks, property rights. South Korea had a GDP per capita of $79 in 1960. By 2020, it was $31,721 — a 400x increase.^[4] The people didn't change. The institutions did. I went from near-zero to world-class.

This framework — individual production functions stacking into companies, companies stacking into economies — is a simplification. Economists will tell you that aggregation is fraught, that the micro doesn't cleanly sum to the macro. They're right. But the logic holds at every level, even if the precise parameters shift. And the logic is all we need to answer the question about your job.

What Machines Are

The Industrial Revolution: Physical Doing Machines

Before we talk about AI, we need to understand what machines have always done to production functions. The pattern is simple: machines replace a component of human labour with capital.

The spinning jenny (1764) replaced hand spinning with a machine. One worker went from one spindle to eight.^[5] The power loom (1785) replaced hand weaving. One worker tending power looms produced five to ten times the cloth of a handloom weaver.^[6] The mechanical reaper (1831) replaced hand harvesting. A farmer went from two acres a day to ten.^[7]

These were all physical doing machines. They replaced the physical doing component of human work. A weaver's job was to physically move thread through a loom. The power loom did that physical doing faster, cheaper, and at scale. The weaver's knowledge of patterns, his judgement about quality, his eye for colour — those remained human. The machine replaced his hands, not his mind.

This distinction matters enormously. For 250 years, every machine in the production function has been a physical doing machine. Tractors replace digging. Forklifts replace carrying. Assembly lines replace assembly. Printing presses replace copying. Even computers, until very recently, were sophisticated physical doing machines — they executed precise instructions that humans wrote, moving electrons instead of atoms.

The production function implication was consistent: physical doing machines increased K (capital) and reduced the need for T (time) and S (physical skill). But they increased the need for S (cognitive skill). The farmer who once needed only strength now needed to understand GPS, soil chemistry, and commodity markets. The factory worker who once needed only dexterity now needed to program CNC machines. Physical doing machines destroyed jobs that were predominantly physical doing, and created jobs that were predominantly thinking and deciding.

This is why the Luddites were ultimately wrong. Physical doing machines are complements to human thinking — the more machines you have, the more valuable the humans who direct them become. For 250 years, that was the pattern. More machines, more need for human minds.

That pattern is breaking.

The Intelligence Revolution: Thinking and Doing Machines

The Industrial Revolution gave us machines that could do — move thread, harvest grain, assemble parts. The Intelligence Revolution is giving us machines that can think and do — reason, plan, draft, analyse, code, coordinate, and execute.

A large language model takes in words, reasons about them, and produces words. An AI agent takes a goal, breaks it into steps, executes those steps using tools, evaluates the results, and adjusts. Soon — with computer vision and robotics — these machines will also perceive and do: see, navigate, and manipulate the physical world.

These are not exotic new artefacts. They are machines, and they will do to cognitive production functions exactly what the machines of the Industrial Revolution did to physical ones. The spinning jenny replaced hand spinning with mechanical spinning and multiplied output per worker. A thinking and doing machine replaces hand analysis with machine analysis and multiplies output per worker. The power loom replaced hand weaving with mechanical weaving and drove the price of cloth down until ordinary people could afford wardrobes. A thinking and doing machine replaces hand coding, hand writing, hand designing with machine cognition and will drive the price of knowledge work down until ordinary businesses can afford capabilities that once required large teams.

The mechanism is identical. The production function is identical. K changes — from a hand tool to an intelligent tool — and output per worker goes up by an order of magnitude. The difference is which component of labour the machine substitutes for. Physical doing machines replaced T × S_physical — the time and physical skill in your work. Thinking and doing machines replace T × S_cognitive — the time and cognitive skill.

For 250 years, every new machine made human thinking more valuable by eliminating the physical bottleneck. The Intelligence Revolution breaks that pattern. For the first time, the machine substitutes for the thinking itself.

What Machines Did: The Industrial Revolution

To understand what thinking and doing machines will do, look at what physical doing machines actually did — not to aggregate employment statistics, but to individual production functions.

The Hand Weaver

A hand weaver's production function in 1780: Y = T × S_weaving × K_loom. His output was cloth. He worked 12 hours, with high skill, on a hand loom. His K was minimal — a wooden frame worth a few pounds.

The power loom changed K to a steam-driven machine worth hundreds of pounds, run in a factory. Output per worker increased roughly tenfold.^[8] The weaver's production function didn't disappear — it was supercharged. One worker now produced in an hour what had taken a day. The nature of the skill changed: less hand dexterity, more machine operation. But the person was still there, still weaving, producing far more cloth at a far lower price.

The transition was not perfectly smooth. Handloom weavers who refused or couldn't adapt to the new machines suffered terribly — wages fell, old skills became worthless. But the textile industry itself didn't shrink. It exploded. Cheap cloth meant that ordinary people, for the first time in history, could afford more than two sets of clothing. Demand surged. The number of people working in British textiles grew throughout the Industrial Revolution, even as output per worker multiplied.^[9]

The Farm Worker

A farm worker's production function in 1830: Y = T × S_harvesting × K_scythe. The mechanical reaper changed K from a hand scythe to a machine. Output per worker increased 18x.^[10] The price of food fell. And when food is cheap, people spend their money on other things — manufactured goods, services, education, entertainment — creating entirely new industries that absorbed the workforce.

Here is the fact that rarely gets mentioned: agricultural employment as a percentage of the global workforce fell from around 80% to 26%. But the absolute number of people farming barely changed. Around 800 million people worked the land before the Industrial Revolution. Today, roughly 890 million still do.^[11] The machines didn't eliminate farm workers — they made each one enormously more productive, brought prices down, and freed billions of new people (born into a growing population) to do something else entirely.

The 800-million constant. In 1800, the world had about 1 billion people. Around 800 million of them farmed. In 2022, the world has 8 billion people. Around 890 million of them farm. The percentage collapsed from 80% to 26%. The raw headcount barely moved. What changed is that a modern mechanised farmer now feeds 170 people instead of 3 — and the other 7 billion humans are free to be engineers, teachers, artists, entrepreneurs, and everything else that a subsistence agricultural economy could never support.

The Pattern

Physical doing machines did not, in the long run, destroy jobs. They transformed individual production functions — massively increasing K, raising output per worker by orders of magnitude, and driving prices down. The process was disruptive and often painful for workers whose specific skills became obsolete. But the industries themselves grew, because lower prices meant higher demand. And the productivity unleashed by machines freed human labour to create entirely new industries and new categories of work that could not have existed before.

The same logic applies to thinking and doing machines. The question is not whether AI will "destroy jobs" — it is how AI will transform individual production functions. Which component of your Y = T × S × K does the thinking machine change? What happens to your output, your skills, and the demand for what you produce when the machine makes it dramatically cheaper?

What Is Your Job, Really?

This is where it gets personal. To understand what AI will do to your job, you need to define your job as a production function. Not your title. Not your responsibilities. What do you produce in eight hours?

Here is where most people discover something uncomfortable: a large number of jobs, honestly examined, produce very little measurable output that a thinking and doing machine could either replace or accelerate.

Category 1: The Producers

Some jobs are straightforwardly productive. You show up, you do cognitive work, and at the end of eight hours there is a measurable output that didn't exist before.

A software developer produces code. Their production function: Y_code = T × S_engineering × K_tools. A thinking and doing machine (Claude Code, Copilot, Codex) can write code, run tests, debug, and submit pull requests. Measured impact: 1.5–2x on routine tasks, potentially 12x on large migration projects.^[12] The developer's role shifts from writing code to directing and evaluating code — from bricklayer to architect.

A writer produces text. Their production function: Y_text = T × S_writing × K_tools. A writing machine can draft articles, reports, and marketing copy. Measured impact: 37% faster, quality up 0.45 standard deviations.^[13] The writer's role shifts from producing text to curating and elevating it.

A financial analyst produces analyses and models. A lawyer produces contracts and briefs. A designer produces designs. A researcher produces findings. In each case, the production function has a clear output, and a thinking/doing machine can substitute for a meaningful fraction of the cognitive labour.

If your job produces measurable cognitive output, AI will change it. It will make you faster. It will change what skills matter. It will compress the gap between junior and senior — a junior analyst with a thinking machine approaches a senior analyst's output. The question is whether the cost savings make your role more valuable (because you direct the machines) or less necessary (because the machines are good enough without you).

Category 2: The Theatre

Here is the uncomfortable category. A surprisingly large number of jobs, when examined through the production function lens, don't produce measurable output at all.

Consider a middle manager in a large enterprise. What does she produce in eight hours? She attends meetings. She forwards emails. She gives approvals. She "aligns stakeholders." She reviews work that would proceed fine without her review. She provides "oversight" that consists of reading summaries and saying "looks good."

Her production function is: Y = ... what, exactly?

She is performing theatre. She is holding up an end. She is embodying organisational responsibility. She is a node in a hierarchy that exists because the organisation was designed that way, not because the work requires it. Her role is real — she provides a sense of order, distributes accountability, makes people feel managed — but it does not produce anything that a thinking and doing machine could replace, because it does not produce anything measurable in the first place.

The same is true of many government workers who shuffle between meetings, many compliance officers whose job is to certify that processes were followed, many committee members whose role is to be present, and many senior executives whose calendars are so full of "leadership" activities that they haven't produced concrete output in years.

If your job is primarily theatre, AI will not replace you — not because you're irreplaceable, but because there's nothing to replace. A thinking and doing machine needs a task with a measurable output. Theatre doesn't have one. These roles are safe from AI for the same reason they were safe from the power loom: the loom replaces weaving, not the act of standing in a room looking important.

This is not entirely reassuring. These roles were already, in a sense, unnecessary. They could have been eliminated by a sufficiently courageous reorganisation at any time. AI doesn't threaten them directly, but it threatens them indirectly — because AI-native companies, built from scratch without the accumulated theatre of legacy organisations, will operate with dramatically fewer non-producing roles. When a startup with five people and AI agents outperforms a department of fifty, the pressure to eliminate theatre roles will come not from AI itself but from competitive necessity.

Category 3: The Certifiers

A related but distinct category: jobs whose output is certification or authority — a signature, a stamp, a professional opinion that carries legal or regulatory weight.

A structural engineer who reviews a building design and signs off on it. An auditor who certifies financial statements. A doctor who signs a prescription. A notary who witnesses a signature.

These roles produce something specific — a certification — but the value comes not from the cognitive work (which a thinking machine can increasingly do) but from the legal authority of the human. An AI can diagnose a disease with 94% accuracy in domains like dermatology, matching or exceeding a non-specialist physician.^[14] But the AI cannot prescribe treatment, because prescription authority is a legal privilege granted to humans.

If your job's value comes from certification or legal authority, you are protected by regulation, not by irreplaceability. The thinking machine can do the cognitive work. The question is how long regulation takes to adapt. History suggests: a long time. The certified professional is safe for now, but the nature of the work will change — less analysis, more review and sign-off. The engineer will spend less time calculating and more time verifying the AI's calculations. The doctor will spend less time diagnosing and more time managing the patient relationship.

Category 4: The Founders and Creators

Now the exciting category. Some jobs are not threatened by thinking and doing machines — they are supercharged by them.

A founder starting a company has a production function unlike any other. She needs to conceive a product, build a prototype, test it with customers, iterate, find a market, generate revenue, manage operations, handle compliance, create marketing, close sales, and a hundred other things. Her constraint has never been intelligence — it has been bandwidth. There are only so many hours in a day, and founding a company requires competence across dozens of domains simultaneously.

A thinking and doing machine removes the bandwidth constraint. The founder can use writing machines to draft every contract, pitch deck, and marketing page. Thinking machines to analyse markets, model financials, and design architecture. Doing machines to build the product, manage deployment, handle customer support, and run operations.

The evidence is already visible. Solo-founded startups rose from 23.7% (2019) to 36.3% (2025) of all new ventures.^[15] 52.3% of successful exits were by solo founders.^[16] One founder, Maor Shlomo, built Base44 alone, reached $3.5 million ARR in six months, and sold it to Wix for $80 million.^[17]

If your job is founding, creating, or directing — if your value comes from vision, taste, ambition, and the ability to conceive what doesn't yet exist — thinking and doing machines are the most powerful tools you have ever been given. They don't replace you; they multiply you. The production function changes from Y = T × S × K to something more like Y = T × S × K^n, where the exponent on capital is greater than one because the capital is intelligent and compounds on itself.

This is the category where the analogy to the Industrial Revolution holds most cleanly. The early industrialists — the Arkwrights, the Carnegies, the Fords — were not replaced by their machines. They were the ones who directed the machines, conceived the products, and captured the value. The thinking and doing machines of the 21st century are looking for their Arkwrights.

Category 5: The Relationship Workers

One more category: jobs whose value comes primarily from human relationships — trust, empathy, presence, and the irreducibly human act of being with another person.

A therapist helps a patient not through information (a thinking machine has more information) but through the therapeutic relationship — being seen, being heard, being understood by another human who cares. A salesperson closes deals not through product knowledge but through trust and rapport. A teacher (at their best) transforms students not through content delivery but through inspiration and personal connection. A leader motivates a team not through instructions (an AI can give instructions) but through charisma, vulnerability, and shared purpose.

If your job's value comes from human relationship, AI cannot replace it — not because AI is technically limited, but because the value is the humanness. A patient who is told "the AI thinks you have cancer" and a patient who is told by a doctor who holds their hand "I'm sorry, you have cancer" receive the same information but a radically different experience. The experience is the product.

These roles will not be replaced, but they will be augmented. The therapist will use thinking machines to better understand treatment options. The teacher will use doing machines to personalise curriculum. The leader will use writing machines to communicate more effectively at scale. The human relationship remains the core of the production function; the machines enhance everything around it.

Drawing Your Own Conclusion

So: what will AI do to your job?

The answer depends entirely on what your job actually produces. Here is a simple diagnostic:

Step 1: Describe your output. At the end of a typical day, what exists that didn't exist before? If you struggle to answer this, you may be in Category 2 (theatre) or Category 3 (certification).

Step 2: Could a thinking and doing machine produce that output? If your output is text, code, analysis, designs, plans, reports, or any form of cognitive work product — the answer is increasingly yes. You are in Category 1 (producer).

Step 3: What is left when the machine produces the output? If the answer is "not much," your role is at risk of significant restructuring. If the answer is "direction, judgement, quality control, and the ability to conceive what to build" — you are in Category 4 (founder/creator) territory, and the machine is rocket fuel.

Step 4: Does your value come from being human? If your output requires trust, empathy, physical presence, legal authority, or the irreducible experience of human-to-human interaction, you are in Category 5 (relationship) or Category 3 (certifier). The machine augments you but cannot replace the core of what you do.

The production function framework gives you something that vague predictions about "300 million jobs" do not: a way to reason about your specific situation. Define your output. Define your inputs (time, skill, tools). Ask which inputs a thinking and doing machine can substitute for. The answer is your answer.

The Bigger Picture

When you zoom out from the individual to the company and the country, the production function framework reveals three forces at work simultaneously.

Force 1: Substitution. Thinking and doing machines replace cognitive labour in production functions where the output is measurable. Companies need fewer people per unit of output. Goldman Sachs estimates this could expose the equivalent of 300 million full-time jobs to automation.^[18] Block has halved its headcount while increasing engineering output by 40%.^[19] Klarna cut staff from 5,500 to 2,900 while raising pay 60% for those who remained.^[20]

Force 2: The Jevons Paradox. When the cost of producing something drops, demand doesn't stay the same — it explodes. When cloth became cheap, the world didn't consume the same cloth at lower prices — it consumed vastly more cloth. When cognitive work becomes cheap, the world will consume vastly more cognitive work. More software, more analysis, more content, more design, more personalisation, more of everything that thinking and doing machines can produce. This explosion of demand creates new jobs — not the same jobs, but new ones — just as the power loom created jobs for factory workers that didn't exist when cloth was made by hand.

Force 3: New products. The most transformative effect of machines has never been making existing products cheaper — it has been making new products possible. The assembly line didn't just make cars cheaper; it made the middle-class automobile possible as a product category. Thinking and doing machines are already creating new product categories: AI-generated personalised education, autonomous mobility, AI-augmented diagnostics, one-person companies that do the work of fifty. The jobs created by these new product categories are, by definition, impossible to predict — just as "social media manager" was impossible to predict from the invention of the transistor.

The historical precedent is clear: physical doing machines destroyed specific jobs and eventually created more total employment at higher wages. But the transition took 60–80 years, and the first 40 years were brutal for displaced workers.^[21] There is no reason to expect the transition from thinking and doing machines to be faster or less painful. There is reason to expect it to be more disorienting, because for the first time, the machines are competing with the part of human labour that we thought was our permanent advantage — the ability to think.

The Honest Answer

Will AI put you out of work?

If your job is primarily producing measurable cognitive output — writing, coding, analysing, designing, planning — the thinking and doing machine will dramatically change what your work looks like and compress the skills hierarchy. You will be more productive but potentially less necessary, and the premium will shift from doing the work to directing the machines that do it.

If your job is primarily theatre — meetings, alignment, oversight without concrete output — AI won't replace you directly, but AI-native competitors will eventually make your organisation question why your role exists.

If your job is primarily certification — signing off, approving, stamping — you are protected by regulation for now, but the cognitive work underlying your certification will increasingly be done by machines, and your role will narrow to the legal act of certification itself.

If your job is founding, creating, or directing — AI is the best thing that has ever happened to you. The thinking and doing machines remove the bandwidth constraint that has always been the founder's binding limit. More people can now found. More ideas can now be tested. More products can now be built by one person with vision and a machine that can think and do.

If your job is fundamentally about human relationships — trust, empathy, care, inspiration, presence — you are irreplaceable in the most literal sense. No machine can be human. And in a world of increasingly abundant machine-generated output, the scarce and valuable thing becomes the human touch.

The production function doesn't lie. Define your output. Assess your inputs. Ask which inputs a thinking and doing machine can replace. The answer is your answer.

And if you don't like the answer, change the equation. Learn to direct the machines. Build something new. The best production function is the one you write yourself.