Bullshit automation is not progress

There’s a lot of excitement right now about Claude Cowork. It’s a step towards agentic behaviour: you give it access to part of your computer, and it can automate work across familiar tools like email, spreadsheets and external services.

But what are we actually automating?

Before we get to value, there’s a background hum of risk we shouldn’t pretend away. Handing AI systems access to your files and workflows assumes nothing goes wrong, nothing hallucinates, and nothing inconvenient happens later when political or commercial pressures shift. History suggests that’s an optimistic assumption.

However, what really bothers me is something else.

I’ve seen at least one influencer celebrating Claude Code because they can generate blog posts for their “community” from a single‑line prompt and existing material. That’s not insight. That’s content recycling at scale. If your contribution is just the highest‑probability answer dressed up as thought leadership, you’re not adding value. You’re creating slop.

The same pattern shows up with spreadsheets, email and PowerPoint. People are excited about automating them, but almost nobody is asking the prior question: is this the right way to meet the need at all?

Spreadsheets are a very old way of analysing information. Email is a dreadful communication and coordination tool. PowerPoint ...argghhh!!! Making these artefacts faster doesn’t mean we’ve chosen the best way to work in the first place.

David Graeber, the anthropologist, coined the idea of “bullshit jobs” to describe work that adds little or no value. He was inundated with messages from people who realised that their jobs — or at least many of their tasks — didn’t really need to exist.

Watching the reaction to Cowork, and to the many other so‑called agentic tools already here or on the way, prompts a mischievous thought experiment.

If today’s AI can automate a task cleanly and easily, is there a good chance that the task never truly needed a human at all, given today’s capabilities? And instead of automating it, should we be asking what need it was trying to serve and how we could better meet that need now, using AI and other tools?

Automating bullshit just gives us faster bullshit.