Practice: Will AI take my job?

Self-check

Six short questions. Answer each in your head before opening the collapsible. Active retrieval is where the learning sticks.

1. Why is “will AI take my job?” the wrong question, and what is the better one?

Show answer

Because you picture your job as a title, but no AI system does a title. It does tasks. It does tasks, the specific pieces of work a job is made of. So “will AI take my job?” has no clean answer, while “which of my tasks are exposed, and which way?” has a very useful one. The move that changes everything is to break the job apart into its tasks and ask what AI does to each one.

2. What does “exposed” mean, and does exposed mean doomed?

Show answer

Exposed means a task is reachable by AI, in either direction: AI could automate it, taking it over, or augment it, helping you do it faster or better while you are still needed. Exposed does not mean doomed. It means reachable, and reachable can be a gift as easily as a threat. Collapsing “exposed” into “job loss” is the single most common mistake, and the whole lesson turns on not making it.

3. What are the three things that can happen to a single task?

Show answer

AI can automate the task, doing it instead of you. It can augment the task, helping you do it faster or better while you are still needed. Or it can create a new task that did not exist before, which someone then has to do. The first two are what “exposed” covers; the third is part of why new work keeps appearing across the whole economy.

4. What decides how much your whole job changes?

Show answer

Two things: how many of your tasks are exposed, and how much of your time those tasks eat. A job made almost entirely of exposed tasks that get automated is at real risk. A job where one small task is automated, or where the exposed tasks get augmented instead, changes rather than ends. Same technology, opposite outcomes; the difference is the mix and the time, not the name on the door.

5. Are we going to run out of jobs?

Show answer

History says probably not, though never painlessly. When automatic teller machines spread, the obvious prediction was the end of the bank teller, but the number of tellers rose for years, because cheaper branches meant more branches. New work keeps appearing because we want new things as old ones get cheaper, most work gets augmented rather than erased, and wholly new kinds of work get invented. The honest crack: earlier machines took physical and routine work and mostly left thinking and language alone, while this wave reaches straight into cognitive and language work, so the shape of work may shift faster than before, and real people still lose specific jobs and have to move.

6. What should you make of the 2023 Goldman number and the 2026 studies?

Show answer

Treat both the scary and the soothing headlines as claims to weigh, not facts to fear. The 2023 Goldman Sachs figure, that generative AI could expose the equivalent of 300 million full-time jobs to automation and eventually raise global output around 7 percent, is a projection made in 2023 about the decade ahead, not a measurement; the bank itself said the work would mostly be complemented rather than replaced and left the numbers out of its own baseline forecast. The 2026 labor evidence is early and genuinely contested: one study found a marked early-career employment drop in exposed jobs, but another 2026 analysis of the same jobs finds no such junior versus senior divide, and the first study’s own authors trace part of the early drop to other, non-AI factors.

Try it yourself

Two exercises, and an optional third in Clawless. The first two you can do on paper.

Exercise 1: Sort the tasks by exposure

Below are eight short task statements. For each, do two things. First, label it exposed or not exposed, where exposed means today’s AI could reach it. Second, if it is exposed, mark whether AI would most likely automate the task (take it over) or augment the person (help them do it faster or better while they are still needed). Then open the key.

Drafting a routine status email to a team.
Comforting an upset client through a hard piece of news.
Reconciling a spreadsheet of transactions against receipts.
Deciding which of two strategies the business should pursue.
Lifting and stocking shelves in a warehouse.
Summarizing a long report into its key points.
Writing the first rough draft of a section of computer code.
Physically caring for a patient who cannot move on their own.

Show answer key

Exposed, mostly automate. A routine status email is language work, the home turf of today’s AI, and a short factual note is the kind of task a machine can largely take over.
Not exposed. This needs a human in the room for reasons we care about: presence, trust, and reading a person in real time. It is human ground.
Exposed, mostly automate. Structured reconciliation against records is repetitive screen work that AI can increasingly do end to end.
Exposed, augment. This is the trap. A model can gather information, lay out options, and argue both sides, which makes this exposed. But exposed is not doomed: the judgment call about which strategy to pursue is exactly the kind of work AI augments rather than replaces. If you read “exposed” here as “a machine will make this decision for you,” you made the mistake the whole lesson is about. Exposed means reachable, and here reachable makes you faster and better informed, not obsolete.
Not exposed. This needs hands in the physical world. Today’s AI reaches language and thinking-on-a-screen work far more easily than physical labor.
Exposed, mostly automate. Summarizing text is close to the center of what these tools do well.
Exposed, could go either way. Generating a rough first draft of code is highly exposed; whether it automates a junior task away or augments an experienced person depends on how much judgment surrounds it. Note how the same technology points in opposite directions depending on the task around it.
Not exposed. Hands-on physical care of a person is human ground, both physically and for the human presence we want in it.

The single most useful habit here is refusing to read “exposed” as “gone.” Exposure only means AI can touch the task; whether that helps you or replaces you is the real question, and it splits case by case.

Exercise 2: Map your own work

This is the tool the whole lesson was built to hand you. Take your own job, or one you care about, and do what the course does.

List your tasks. Write down 6 to 10 of the tasks your job is actually made of, the way you would explain your week to a friend. Not “being an accountant,” but the pieces: reconciling the books, explaining a number to a nervous client, chasing a missing document, deciding what a strange figure really means.
Mark each one exposed or not. Beside each task, ask: could today’s AI do a real part of this? Language and thinking-on-a-screen tasks are usually exposed; tasks that need hands in the physical world, or a human in the room for reasons you care about, usually are not.
For each exposed task, ask automate or augment. Would AI take the task over, or would it help you do it faster and better while you are still needed and freed for the human work around it? Be honest in both directions; do not read every exposed task as a threat, and do not wave every one away.
Weigh the mix by time. Roughly, how much of your week do the exposed-and-automated tasks eat, versus the augmented ones and the human ones? A job whose time-heavy tasks get automated changes far more than one where only a minor task does.
Name your solid ground. Circle the human tasks: the judgment, the relationships, the care, the parts people want a person for. That is where your ground is most solid, and where to keep growing.

You will almost never find a job that is all one thing. You will find a mix, and the mix is your map. This is the same lens you met in Lesson 4, where you asked whether a task should be handed to AI at all; here you turn it on your own living.

Exercise 3 (optional): Talk it through in Clawless

This step runs in Clawless, the working environment we use across Clawdemy, and a plain conversation is all it needs. A model is not an oracle about your real career, so do not ask it to tell you whether your job is safe. Instead, use it to sharpen the thinking you did above:

Ask the model to help you brainstorm the full task list for your job, in case you left tasks out. You then do the exposure scoring yourself, because you are the one who knows which tasks really matter and how your time is spent.
Ask it to role-play a friendly skeptic who keeps pushing you to defend which of your tasks are truly human ground. Say your reasons out loud, in your own words. The goal is to test your own map, not to get a verdict.

Keep it to plain conversation. The point is to make your own map sharper, not to hand the decision to a machine. A model can be a good sounding board; it is not an oracle about your real career, and the person who can map how AI touches their own work is already more powerful than the one who fears it in the abstract.

Flashcards

Q. What is the core move: is a job one thing, or many?

A job is a bundle of tasks. When you worry about your job you picture a title, but no AI system does a title. It does tasks, the specific pieces of work a job is made of. A bookkeeper does not do one thing called bookkeeping; she enters transactions, chases receipts, reconciles accounts, and reassures a worried owner, all different tasks that AI does not treat alike. So the frightening question becomes workable: break the job apart and ask what AI does to each task.

Q. What are the three things that can happen to a single task?

AI can automate the task, doing it instead of you. It can augment the task, helping you do it faster or better while you are still needed. Or it can invent a new task that did not exist before, which someone then has to do. The first two are what makes a task exposed; the third is part of why new work keeps appearing across the whole economy.

Q. What does exposed mean, and does exposed mean doomed?

Exposed means a task is reachable by AI, in either direction: AI could automate it, taking it over, or augment it, helping you do it faster or better while you are still needed. Exposed does not mean doomed. It means reachable, and reachable can be a gift as easily as a threat. Collapsing exposed into job loss is the mistake almost everyone makes, and the whole lesson turns on not making it.

Q. What decides how much your whole job changes?

Two things: how many of your tasks are exposed, and how much of your time those tasks eat. A job made almost entirely of exposed tasks that get automated is at real risk. A job where one small task is automated, or where the exposed tasks get augmented instead, changes rather than ends. The difference is the mix and the time, not the name on the door.

Q. Which tasks are most exposed to AI today?

Today’s AI is strongest at language and thinking-on-a-screen work: writing, summarizing, drafting, coding, analyzing. So a task that is mostly words, or mostly reasoning at a keyboard, is highly exposed. A task that needs hands in the physical world, or a human in the room for reasons we care about, is much less so. Words are these systems’ home turf, because underneath they are prediction engines trained on oceans of human text.

Q. Which is more exposed, a software developer or a human resources manager?

The software developer, by a clear margin, from a World Economic Forum analysis the course cites. So much of software work is text a machine can now generate, which makes it highly exposed. Human resources leans harder on judgment, persuasion, and sitting with people through hard conversations, which today’s tools reach less easily. But watch the trap: more exposed is not more doomed, because exposure runs both ways.

Q. Are we going to run out of jobs?

History says probably not, though never painlessly. When automatic teller machines spread, the prediction was the end of the bank teller, but the number of tellers rose for years, because cheaper branches meant more branches. Telephone operators, by contrast, faded away. One technology can destroy some work, transform other work, and invent work nobody imagined, all at once.

Q. Why does new work keep appearing?

The labor economist David Autor, whom the course follows, points to three forces. As old things get cheaper, we want new ones. Most work gets augmented rather than erased, so the remaining human parts gain value. And wholly new kinds of work keep being invented, which is why so many of today’s jobs did not exist a few generations ago.

Q. Where does the comfort from history crack?

The old machines took over physical and routine work and mostly left thinking and language alone. This one reaches straight into cognitive and language tasks, the very ground white-collar workers stood on while the factories automated. So the comfort from history is partial: we are likely to still have work, but its shape may shift faster, and reach further up the ladder, than before. And even in the good case, real people lose specific jobs and have to move.

Q. What should you make of the 2023 Goldman Sachs number?

Handle it as the forecast it is. In 2023, Goldman Sachs estimated that generative AI could expose the equivalent of 300 million full-time jobs to automation and could eventually raise the world’s yearly output by around 7 percent. That is a projection made in 2023 about the decade ahead, not a measurement of anything that happened. The bank itself said the exposed work would mostly be complemented rather than replaced, and was uncertain enough to leave the numbers out of its own baseline forecast. Even the alarming headline was really saying that a great deal of work is reachable, and reachable is not the same as gone.

Q. What is Baumol's cost disease, in one line?

Some work never gets more productive, where productivity just means how much you get out for the effort you put in: it still takes four players just as long to perform a string quartet as it did two centuries ago. Yet those players earn more, because they must be paid enough to keep them from other jobs. So the corners of the economy hardest to automate, often human ones like care and teaching, grow as a share of what we spend and drag the whole speed-up down. Unless AI lifts every corner, the stubborn ones set the pace.

Q. What is Solow's timing lag, in one line?

When computers first flooded into offices, the economist Robert Solow observed that you could see the computer age everywhere except in the productivity numbers. The gains took years to show up, because people and firms had to rebuild how they worked around the new tool first. AI may follow the same curve: much activity and investment now, with the real payoff, and the real disruption, arriving later than either the boosters or the doomers expect.

Q. Define task, automation, augmentation, exposure, and productivity.

A task is one of the specific pieces of work a job is made of. Automation is AI doing a task instead of the person. Augmentation is AI helping a person do a task faster or better while they are still needed. Exposure means a task is reachable by AI in either of those two ways, which is not the same as job loss. Productivity is how much you get out for the effort you put in.