There's an office scene that feels all too familiar to me. Someone prepares a report for a meeting. In the meeting, they agree to ask for a shorter version. That version becomes a presentation nobody has time to read before the next meeting. Everyone was busy. It's much less clear what changed.
Now imagine an AI producing the report, the summary, and the slides in minutes. Did we solve the problem? Or did we just make the loop run faster?
That question brought me back to David Graeber's Bullshit Jobs. The title is provocative, even a little insulting at first glance. But the provocation isn't aimed at working people. It's aimed at a way of organizing work that asks them to spend their time, intelligence, and part of their lives on something whose purpose even they struggle to defend.
First, let's not confuse two things
Graeber distinguished between work that's hard, badly paid, or poorly treated and a job the person doing it feels shouldn't exist, even though they have to act as if it matters. Cleaning a hospital can be exhausting and unfairly paid; that doesn't make it useless. On the contrary, its value becomes obvious when nobody does it.
Nor can we look at one isolated task and declare the person doing it unnecessary. A form might look absurd until you learn that it creates a record needed to protect someone. And a genuinely pointless task might be only one part of an otherwise valuable job.
Graeber's claim about how widespread meaningless jobs are has been challenged by researchers. Other studies find partial support while showing that working conditions matter too. We don't need to settle that debate here to recognize something closer to home: most of us have seen processes survive because “we've always done it this way,” because someone needs to show activity, or because nobody has permission to ask whether they still serve a purpose.
The easy promise: let a machine do it
AI looks well suited to some repetitive work: copying data between systems, drafting standard documents, sorting requests, finding scattered information. If those tasks are necessary but tedious, taking them off someone's plate could genuinely improve their day.
But there's a trap. Automating a task doesn't prove it should exist.
If the report nobody uses now takes thirty seconds to generate, it's still a report nobody uses. Worse, because it's cheaper to produce, we may start requesting twenty of them. AI can free a person from a routine, or it can create a new layer of emails, metrics, and meetings to oversee its own output.
I catch myself doing this too. When a tool offers to make something quickly, it's tempting to accept before asking whether that thing helps. Speed is seductive. Meaning takes a pause.
The question before the prompt
Before automating a process, I'd start with questions that are much less technological:
- Who uses this result, and what decision does it help them make?
- What would happen if we stopped producing it for a month?
- Which part protects someone or meets a real obligation, even if that isn't obvious from the outside?
- Which part exists only to show that we were busy?
- If we save time, will the person gain autonomy and room to learn, or simply receive more tasks?
The people doing the work should help answer those questions. They know the shortcuts, the exceptions, and why a procedure that looks absurd might still be there. They also know how much it weighs on them to pretend to care about it.
That knowledge doesn't fit neatly into a flowchart or a chatbot history.
The opportunity isn't to erase people
It would be cruel to use Graeber's idea to tell someone, “Your job doesn't matter, so an AI can replace you.” That person didn't create the role alone, relies on its income, and may be doing valuable work the formal job description doesn't even mention.
A more interesting opportunity would be to remove bureaucracy without removing dignity: redesign processes with the team, stop demanding deliverables for show, direct capacity toward real needs, and share the benefit of the time saved. In some places that could mean better service. In others, more room to learn, care, or simply finish the day without being drained.
None of that comes built into the model. Those are organizational decisions and, ultimately, questions of power. AI can make it painfully clear which tasks we know how to automate. It can't decide for us what deserves to exist or how to treat the people caught in the transition fairly.
Maybe that's why the question that interests me most isn't “How many jobs can AI eliminate?” It's this: if we can stop doing some of the empty work, will we choose to stop, or will we use technology to produce it faster?
I explored this idea some time ago in a video on my channel. The question feels even more urgent to me now. I'd love to know which tasks in your own work you'd put on the table before trying to automate them.