top of page

Tech & Society

What 81,000 real AI conversations reveal about how we actually use it

Anthropic analysed 81,000 real Claude conversations to map actual AI usage at work — finding it concentrated in writing, coding, and research, and far more about augmenting tasks than replacing them.

Everyone's been arguing about whether AI will take your job. Turns out most people are just using it to fix their emails.

 

Anthropic's dataset is rare because it shows behaviour, not intentions. No surveys, no "how do you feel about AI at work" but 81,000 conversations. What people report to the AI bot they do, overwhelmingly, is ask for help with writing, coding, and summarising things. Not autonomous agents running entire workflows.

 

Every technology that eventually rewired how we work entered through the same door: low-risk, easy-to-reverse tasks. The spreadsheet didn't arrive as a strategic transformation — it arrived as a slightly better way to do what accountants were already doing. Then it quietly changed what was expected of every accountant.

 

The more structurally telling finding is how uneven adoption is. Knowledge workers are being reshaped first — and fast. Everyone else, largely untouched. The "AI revolution" isn't sweeping through organisations; it's pooling in specific roles and functions, accumulating quietly.

 

But once enough micro-uses stack up, they shift the baseline. What used to take a day takes an hour. And suddenly the expectation recalibrates, but because enough small recalibrations already happened. First quietly, then all at once.



00:00 / 01:08
bottom of page