AI is going to hit like a train.

I've been talking about AI and jobs for years now. On stage, in bootcamps, in this letter.

And for years, I've said the same thing: "AI will take 99% of jobs by 2030" is nonsense.

I still think so.

AI won't take 99% of the jobs we have today. It will take even fewer of the jobs that will exist in 2030, because a lot of them haven't been invented yet. And I don't believe AI can replace humans everywhere, in absolute terms.

But this week I want to be more precise, because this is the kind of idea that hurts you if you hear it wrong.

Many jobs, in the form they have today, are already over.

The people doing them will (need to) find other work. New jobs will appear, as they always have.

(We're not running out of problems to solve).

But a job written as a list of tasks, executed in front of a computer, is losing its reason to exist. And it's going faster than I expected.

If your job is mostly ticking boxes, AI is going to hit like a train.

What history actually says

The record is reassuring in one way and brutal in another.

James Bessen, an economist at Boston University, went through the 270 occupations listed in the 1950 US census. By 2010, exactly one had disappeared mainly because of automation: the elevator operator.

Meanwhile, new work keeps appearing. David Autor and his colleagues at MIT found that around 60% of Americans employed in 2018 worked in job titles that didn't exist in 1940.

So when someone tells you 99% of jobs will vanish, history doesn't back them up.

Then comes the brutal part.

A century ago, telephone operator was one of the most common jobs for young American women. Then mechanical switching took over the calls, and most of those jobs went away.

Two economists, James Feigenbaum and Daniel Gross, followed what happened to the people. The next generation of young women was fine. They went into other work: typists, secretaries, and also lower-paid service jobs. The women already sitting at the switchboard paid the price. Ten years later, they were more likely to be in lower-paid work, or out of the workforce entirely.

That's the part people skip when they say "new jobs will appear."

They will. The real question is whether you're still sitting in the old one when it goes.

What I see in my own company

What sharpened my view this year came from my own week.

We run live sessions with guests in our community. Each one used to be a small project, handled by hand, step by step.

Today my agents do all of it. They create the Zoom link, write the event in the community, draft every email in our voice and schedule it, publish the announcement, promote it on social media and set up the reminders. Same for the podcast guests.

After the live, they take the recording, edit it in Descript against a written definition of what excellent work looks like for us, and prepare the replay in our brand voice.

Our IT is managed by agents too.

(I know how this sounds coming from me. I've spent years being the guy who says "calm down" about AI. That's exactly why I'm telling you.)

What stays with me is a short list. Choosing the guest. Deciding what the session is about. Every invitation waits for my explicit GO before it leaves, and I press publish on every replay.

So when I look at my company honestly, I see two kinds of work. The tasks, which agents now do well. And the will to move things forward, which they don't have at all.

Apart from the people who bring that will, who take full ownership of making something happen or solving a real problem, I could run my entire company with agents.

A few years ago, I would have laughed at that sentence.

How a job becomes a list of tasks

In 1930, John Maynard Keynes looked a hundred years ahead, to 2030, and imagined his grandchildren working fifteen hours a week. Technology would do the rest.

The technology came. The fifteen-hour week didn't.

In 2013, the anthropologist David Graeber wrote an essay asking why. His answer was provocative: we filled the time with jobs that don't need to exist. He called them bullshit jobs. Two years later, YouGov asked British workers whether their job made a meaningful contribution to the world. 37% said no.

Then researchers went looking for those jobs in harder data, and the picture got more interesting. Using a large European survey, a team from Cambridge and Birmingham found that only about 5% of workers felt their work was rarely or never useful. And the feeling followed management more than job titles. People who felt respected and encouraged by their managers were far less likely to call their work useless.

That matches what I see in companies. Most of the jobs I'm talking about do produce something. What they're missing is room for anyone to own the outcome.

They grow in a predictable way, especially in large companies and administrations.

A company grows. Coordination gets harder. So you add a layer to track the work, then a layer to report on the tracking. Processes get written for a problem that disappeared years ago, and nobody removes them. Roles get defined as a list of tasks, because tasks are easy to measure and outcomes are hard.

Little by little, the job becomes: you asked me to do X, Y and Z. I did X, Y and Z.

The person can be smart, kind and hard-working. The organisation simply never asked them for more than execution.

(To be clear: I'm not blaming the people in those roles. The system built the role. But the system won't protect it either.)

For decades, execution in front of a computer was scarce enough to pay a salary. That's what just changed. A job written as a list of tasks is already half a brief for an agent.

Where it already shows

This is starting to show up in the data, in a very specific place.

Across the whole economy, nothing dramatic yet. The Yale Budget Lab has tracked the US labour market since ChatGPT came out, and it still finds no clear sign of disruption from AI.

Look closer, and something moves. A Stanford team led by Erik Brynjolfsson follows payroll data for millions of American workers. In the jobs most exposed to AI, employment for people aged 22 to 25 has fallen about 19% behind young workers in less exposed jobs. Experienced workers in the same jobs show no such gap. And the gap comes mostly from companies hiring fewer young people, much more than from layoffs.

The researchers are careful about cause, and I'll be careful too. But the pattern fits what I see. The first job in a career is usually where the tasks live. You learn by executing. And that's the layer agents absorb first.

What the data shows is a door closing quietly, at the exact place where careers used to start.

What a company actually pays for

So what are you paid for, once the tasks are cheap?

A company pays for sources of will.

Someone who wants the thing to happen. Who sees what's broken and decides it's theirs to fix. Who owns the outcome, including when it goes wrong. Who keeps going until it's done, with AI or without it.

Agents are getting good at spotting problems. Point one at a client file and it may well tell you the numbers have been dropping for three months. It can draft the email, find the CFO's name and book the call.

What it won't do is decide that this client is its problem, and refuse to let go until the problem is solved. An agent works inside the scope someone gave it. It doesn't care whether the client stays. It can't want the result, and it can't own it.

Nobody can hand an agent a broken launch at 9pm and make it answer for the result.

That gap is where your salary lives.

Picture someone who sees the same drop in a client's numbers. The call with the CFO goes nowhere. So they ask a former colleague for an introduction to the COO. They convince their own boss to change the offer. They come back to it every week for two months. And if the client leaves anyway, they're the one who stands up and explains why.

None of those steps was on a task list.

AI can already do the analysis, the email and the booking. It can't make the outcome its own goal and go after it with every means available. The person who does is the one a company can't do without.

Your boss is now your benchmark

Being good with AI is no longer optional. At the very least, you need to be as good with it as your boss.

From your manager's chair, the math is simple. If they can get your deliverable from an agent in ten minutes, faster and cleaner than from you, your job just became a line on their to-do list.

Agents also stretch how much one person can oversee. Someone who knows what good looks like can brief and check a lot of work, with written SOPs (step-by-step instructions an agent follows) and a few clear gates. That's how my own setup runs. The layers whose job was to pass the work along get thinner.

This is why I keep pushing you to build your own system (Letter 105) and to write your context down (Letter 99). The people who do it become more valuable with every model release. The people who wait to be trained become easier to replace with every release.

The honest part

I don't want to sell you a disaster movie.

Companies move slowly. Most processes are still a mess, and you cannot automate a mess (Letter 78). Many IT and security teams are blocking unvalidated AI tools, and they're right to. Contracts, regulation and habit will slow all of this down.

And new jobs will come. Some of them will be built around exactly what agents lack.

What companies need now is drive. Leadership. A sense of entrepreneurship, even inside a big organisation. And if I'm honest, that's rarely how job descriptions are written, or what most people were hired for.

Nobody is born a source of will. It's a habit. You take one thing, you own it end to end, you get it done, and then you do it again with something bigger. It's the same logic as the brain gym (Letter 86): the muscle grows when you use it, and it fades when you hand everything off.

A quick test for your own job

You don't need a consultant to know which side you're on. Answer three questions honestly.

If you stopped for two weeks, what would break? If the honest answer is "a few reports would be late," that's a warning. If the answer is "a client would leave, a launch would slip, a team would get stuck," you're carrying something real.

Can you name the outcome you own? One sentence, a result, with your name on it. "I make sure our biggest clients renew." "I make sure the product ships on time." If all you can name is activities (meetings, updates, tickets), your job is written as tasks, and tasks are exactly what agents are getting good at.

When did you last start something nobody asked you for? A fix, an idea, a conversation that moved a decision. If you can't remember, the organisation has trained you to wait. And waiting is the one thing agents already do perfectly.

What to do on Monday

Three things. You can start all of them on your own.

Rewrite your job as outcomes. Translate every task in your job description into the result it's supposed to produce. "Update the dashboard" becomes "leadership makes the right call on Monday." Then ask which of those results you actually own. Every result you can't claim is your risk.

Solve one problem nobody asked you to solve. Pick something broken that everyone works around. Fix it this month, with AI, and say what you fixed and why. That's what being a source of will looks like from the outside.

Do your own tasks with agents first. Pick the task you repeat every week, write down how you do it, and hand it to an agent your company has approved. Check the result, then move that time to the part only you can do. If your boss is better at this than you, close that gap now.

If you want to see how I work

On Tuesday, I'm running a free live session: "Build your AI Operating System."

I'll show you my actual setup. The folder, the agents, the SOPs they follow, and the gates where they stop and wait for me. You'll see exactly which part of my work moved to the agents, and which part stays with me.

Tuesday 10 November, 18:00 Paris time (17:00 UTC), 60 minutes. It's free, and there's a replay if you can't make it live.

Save your seat →

Back to 2030

Keynes picked 2030 for the fifteen-hour week. The people announcing the end of 99% of jobs picked the same year.

I think both will be wrong about the number.

What I'm sure of is smaller, and much closer to your desk. By 2030, a job that is only a list of tasks will be very hard to defend.

The train is coming for the tasks. Make sure your job is more than that, and make sure the "more" is you.

AI is only as good as the human operating it.

Have a great weekend.

Stay sharp.

— Charafeddine (CM)