In his book Bullshit Jobs, the late, great David Graeber defined bullshit jobs as paid employment so completely pointless and unnecessary that even the employees cannot justify their existence. He identified five types:

Flunkies exist to make somebody else feel important. Think doormen at buildings that don’t need them, receptionists with nothing to receive.

Goons are jobs that only exist because competitors have them. Lobbyists, corporate lawyers, telemarketers, PR specialists—roles locked in an arms race of mutual uselessness.

Duct tapers fix problems that shouldn’t exist in the first place. Their entire job is patching over failures that better systems would eliminate.

Box tickers create the appearance of doing something useful. Compliance officers, survey administrators, anyone whose output is primarily paperwork proving that paperwork was done.

Taskmasters manage other people, often by creating bullshit tasks for them to complete. Middle management, “leadership.”

A lot of people hate this theory, but I still think it’s one of the most underrated sociological frameworks of the 20th century. Look inside any company. If you’ve been in a workplace even for a few years, you’ll be surprised at the amount of pointless nonsense that passes for work—tasks that could be automated or eliminated entirely. But these jobs persist for reasons that even my people-brain can’t fully comprehend.

Now we’re entering the age of generative AI. I don’t know whether there will be an AI-led job apocalypse, but what I am near-certain about is this: there’s going to be a bullshit job apocalypse.

Think about all the jobs that are basically: find this file, tick this box, send this email, copy this number from one spreadsheet to another. Tasks that require no judgment, no creativity, no actual thinking. Decorative employment. Gone.

Now, a counterpoint. Ben Thompson recently wrote a brilliant post in which he noted that humans have always created new jobs to replace the ones that disappeared. He pointed to agriculture: in the 1800s, 80% of the US population worked on farms. By 2000, it was 1%. And yet humans were alright. New kinds of work emerged.

He also made another argument I haven’t been able to stop thinking about: humans want other humans. Even if robots get good at forecasting, art, music, whatever—humans will ultimately value humans over machines.

That’s because humans didn’t just sit on their hands; rather, entirely new kinds of work were created, which were valued dramatically higher. Much of this was in factories, and then, over the last century, there was the rise of office work. All of that could very well be replaced by AI, but the point is that the history of humans is the continual creation of new jobs to be done — jobs that couldn’t have been conceived of before they were obvious, and which pay dramatically more than whatever baseline existed before technological change.

It’s possible, of course — and to return to my perhaps self-interested and potentially misplaced optimism above — that robots will be better at podcasting than Patel or I. I’m skeptical, though: my experience — and I’ll only speak for myself here — is that the human element is essential in creating compelling content. Sure, sometimes I say “uhm” or “like” or “sort of”, or I get facts wrong, but that’s a feature, not a bug: what I have to say is by definition unique to me, and that is interesting precisely because I am flesh-and-blood, not a robot

This could all be true. It could also be wrong. I don’t know.

What I do know is this: the disappearance of a large chunk of white-collar jobs is, in my head, now a given. Whether that leads to some other situation where humans become artisanal, handmade goods—remains to be seen.

I can easily imagine a future where a receptionist, a stenographer, an executive assistant becomes a luxury job. Not a necessity, but a status symbol. An anachronism you keep around because you can afford to. The same might apply to other roles. Human receptionists as live art inside companies, like live music in a hotel lobby. Decorative in a different way.

But that’s speculation for another time.

The reason I’m saying this: today, Anthropic launched Claude Cowork.

Cowork is essentially a normie-friendly version of Claude Code. The problem with Claude Code was that the name was a misnomer. Even though it was originally a command-line interface meant for coders, people ended up using it for all sorts of non-coding tasks. In my own case, I used it to analyze data and create charts for blog posts. I used it to write, edit, format, and clean up posts. I used it to organize my Obsidian notes, merge them, clean them up. I was basically using it like a mini butler—my own Alfred, if you will.

I think even the people at Anthropic were surprised by the use cases. If you go to Claude’s Twitter account, they do a weekly thread of weird and fun things people are building with Claude Code, and the range is remarkable. So they figured it out and built Cowork.

Here’s what it can do: you give it access to a folder, and it can read, write, create, and edit files—documents, presentations, whatever. But it’s not just a cute little virtual container working in a local folder. Cowork has external connectors that can plug into your calendar, your inbox, and various other services. It can actually reach into the systems where real work happens.

And here’s the thing that stopped me cold: Boris Cherny, the creator of Claude Code, confirmed on Twitter that Cowork was pretty much entirely written by Claude Code. The tool built the tool. Brave new world.

Think about what bullshit jobs actually consist of: take this file, edit this, change that, delete that, send this, remind someone about that. Cowork can now do all of it. You tell it what to do, step away from your computer, and it creates a plan, analyzes everything, fixes its own mistakes, and comes back to tell you the job is done.

This is now available to everyone. Which means bullshit jobs are pretty much dead.

Cowork can do a vast chunk of what passes for routine managerial and clerical work in organizations. It’s like having your own assistant—one that’s more competent and more intelligent than most. For all those people who call themselves “analysts,” game over.

One more thing happened today: Apple signed an agreement with Google to embed their AI capabilities into Siri. The reason I mention this is because AI capabilities are now being embedded into devices used not just for personal tasks, but for professional work as well.

What this means: slowly but surely, parts of the job bundle will be unbundled and increasingly delegated to AI agents, assistants, whatever you want to call them.

The likely prospect is not mass unemployment. It’s the steady degradation of employment opportunities. The slow thinning of the job bundle. The progressive merging of all these basic bullshit tasks into tools that do them better, faster, and without complaint.

You can hear the creaking noises. AI is chipping away at large chunks of tasks that most white-collar workers and knowledge workers do. We’re not talking about full automation. We’re seeing the entry-level stuff go first—the basic tasks, the menial things, the manual drudgery involved in knowledge work. That’s being slowly taken over.

But you can also see these tools starting to do more. Encroaching upon things that even six months ago, people confidently said AI couldn’t do. “Your judgment still remains supreme,” they said. That confidence is starting to look misplaced.

So yeah. Pay attention to the creaking noises and the sparks.