#OpenAI Yesterday released a study on how people use AI for work. After reading it, the most striking part to me isn’t “whether AI will replace jobs,” but the fact that many people have already started sneaking into work that crosses roles.

The study’s sample comes from more than 800k job-related messages from US users, and 16.8% of them involve cross-occupation tasks; after removing general work like writing and summarizing, 43.5% of occupation-related messages are actually doing things that other roles would normally handle. Customer support starts getting involved in data analysis, sales checks customer data themselves, and marketing conveniently troubleshoots website issues on the side. In the past, it wasn’t that people didn’t want to do many things—it was that their skills were stuck; they had to find someone, wait in line, and join meetings. Now you hand it to AI to try first, and many things that were beyond an individual’s capabilities can still be pulled off.
This shift is even more obvious in small teams: in 2–5 person workspaces, cross-occupation tasks account for 18.9%, while in workspaces with 100 people or more it’s 16.3%. Small companies were already short on headcount; when AI shows up, the first thing that happens isn’t firing anyone—it’s that one person starts covering two or three roles.
I’ve personally felt this very strongly lately. In the past, a lot of automation, scripts, and tools—things I wanted to do but couldn’t write—could only stay as ideas. Now my Codex and Claude quotas aren’t enough just for everyday use: in the daytime I study work workflows, and at night I keep tinkering with automation, to the point where I almost want to treat one person as eighty donkeys. It’s an exaggeration, sure, but productivity really has been pushed to a new height.
When I trade, I’ve never liked focusing only on big phrases like “will the job disappear.” The market often reacts to more granular things first: can one person take on an extra half role, and can a team still charge in the same way as before? The upside is efficiency goes up, but the downside is also very real—boundaries blur, and a boss may simply assume you can do anything.
So my thought is simple: first, treat AI as a plug-in, and make the kinds of tasks I can take on as wide as possible. As for whether I’ll be replaced, I’ll say what I said before—capital is always about cost-effectiveness!
#Codex #Claude