The latest research from Anthropic collecting interview data from 81,000 people is wild. But I can’t help but think what’s needed is more ethnographic work to truly understand what’s happening with AI day-to-day.
What immediately jumped out at me in the Anthropic work is that people reported “productivity” as the biggest thing that AI has “delivered on.” At the same time, I’ve been personally fixated on the ethnographic research by Xingqi Maggie Ye and Aruna Ranganathan.
This is perhaps the most resonant research on AI that I’ve come across.
They found that AI in the workplace was “intensifying” work rather than streamlining it. Basically, people were expanding job scope, taking on more, working longer hours, and potentially burning out. People also didn’t quite realize this until they reflected on things afterward.
I find that with AI, whether it’s for work or learning, it can often appear to be effective. It can feel good in the moment. It can simulate thinking or ideation or knowledge acquisition, but when you step back it’s like a fever dream.
Afterward, when you account for the time a task took, and perhaps revisit that task via other means later, you might in fact realize you’re better off doing that thing yourself.
Of course this isn’t always the case, and there are things AI can help with, but it often requires a lot of restraint and structure. Working well with AI might mean thinking about a task, deciding when AI is going to play a part, and isolating it to very specific roles. This might be for efficiency’s sake, to lean into what AI is technically best at, or simply to protect your autonomy.
Basically, knowing when not to use AI (and why) is as important as when to use it (and how).
I don’t think many people, myself included, are fully there yet. In such a short time with GenAI, and with such little causal evidence, who really is?
Originally published on LinkedIn on March 19, 2026.