AI Replaces Tasks, Not Jobs
Image: AI-generated
Meet Jean-Claude, the colleague who loves the work the rest of us keep postponing.
Monday, eight in the morning, at the office. My inbox holds more unread emails than the Atlantic holds fish. My to-do list has more entries than Gone with the Wind has pages. My motivation has already hit the floor, and I'd like to go home.
It doesn't help. The work has to be done. But where to start? Prioritising all of it would take a week on its own. And so much of it is dull. Analyse this unstructured data for patterns. Write a report. Run these ten steps a hundred times, identically, then write a manual on how to do it.
This is what I went to school for? What I studied for, what I spent years of knowledge and experience on? And on top of it all, I am expected to move the company forward — to think strategically, to be creative. When, exactly, is never mentioned.
The empty calendar
In a 2017 Bloomberg interview, Bill Gates talks about Warren Buffett's calendar. It is almost empty. Gates admits that he used to have every minute packed and thought that was the only way to get things done. Buffett's point is simpler: he can buy nearly anything, but not more time.
The lesson is not that wealthy men take long lunches. It is that thinking needs room. Whoever wants to be strategic or creative has to be free of distraction and routine first. Leonardo da Vinci did not paint the Mona Lisa between a report on the grain stocks in the warehouses of Florence and ten quotes for portraits to the royal courts of Europe.
Meet Jean-Claude
But wait. I have a new colleague. Let's call him Jean-Claude.
Jean-Claude likes to work, and he likes precisely the work I don't. He is delighted to wade through endless emails, unsubscribe from newsletters, write me summaries and delete everything that doesn't matter. He writes a flawless report. He finds the needle in a haystack that, by any honest estimate, never contained one.
What he is not good at: planning strategically into the blue. Being creative without a nudge.
AI is strategic when we tell it to be and set the frame. It is creative when we give it the brief — write a novel set in the world of The Lord of the Rings, in the language of Shakespeare. And somewhere in writing that brief, we fail to notice that the strategic and creative thinking was ours, not the machine's.
Tasks, not jobs
That is why I am convinced AI does not take our jobs. It takes our tasks — and mostly the ones hardly any of us ever liked. They feel less like work than penance, and we push them ahead of us from one Monday to the next.
Yes, jobs will disappear or change at the root. They always have. They always will. The court jester of old is today's comedian. Or, on some days, today's politician. In return, new jobs appeared for needs nobody had before. A full-stack engineer would have struggled to find work two hundred years ago. So would a radiologist.
Which leaves an open question: what happens to the people whose job was mostly the routine — starting with those at the very beginning? That is the next part of this series.
So let's build them
Let's get going, then. Let's build these helpers, empty our calendars and start on the problems that are actually ours to solve.
And if AI takes over even more than that, John Maynard Keynes may be proven right after all. In his 1930 essay Economic Possibilities for our Grandchildren, he predicted that a hundred years on, we would work three-hour shifts — a fifteen-hour week. That gives us until 2030. My claim: we have never been this close.
Bonus: put Jean-Claude on your inbox
Start where Monday hurts most. What you need is an AI assistant with access to your mailbox — Claude with the Gmail or Outlook connector, Copilot in Outlook, or whatever assistant you are allowed to use. At work, that last part matters: use only tools your employer has approved. A mailbox is full of data that is not yours to hand out.
Then teach it, the way you would teach a new colleague. On the first run, let it go through the last week and propose its own rules: which senders are newsletters, which are noise, which always matter. Correct it where it is wrong. That correction is the training, and it takes an afternoon, not a project.
After that, run the prompt below every morning — or set it up as a scheduled task, so the briefing is waiting for you at eight. One rule stays in place for the first weeks at least: Jean-Claude suggests, you decide. He does not send, delete or unsubscribe anything on his own.
You are my email assistant. Go through all unread emails in my inbox from the last 24 hours.
Sort every email into exactly one group:
1. Act today: needs a decision or a reply from me today.
2. This week: needs a reply, but not today.
3. Read later: useful information, no action needed.
4. Newsletters and promotions.
5. Noise: notifications, confirmations, automated messages.
For groups 1 and 2: one line per email with sender, subject and what exactly is expected of me.
For group 1: draft a short reply to each email, in the language of the email. Do not send it.
For group 4: tell me which newsletters I have not opened in three months and propose unsubscribing.
For group 5: propose what can be archived.
Do not send, delete, archive or unsubscribe anything without my explicit confirmation.
End with three lines: what matters most today, what can wait, and what I can ignore.
Jean-Claude has read this post, by the way. He found two typos and no strategy.
Part one of After the Agents, a series on what work looks like once agents take the tasks. It follows From Excel to AI Agents.