AI isn't taking your job -- Yet!

AI Isn’t Taking Your Job—Yet Most research on AI and employment starts with a simple assumption: if a model can do a task faster than a human, that job is “exposed.” Sounds reasonable, until reality hits. Tasks are messy. Companies are slow. Risk is high. Software stacks are missing. Humans still sign off on everything. Just because AI can do it doesn’t mean it does. Anthropic’s new study tackles this gap. They aren’t saying “AI is taking jobs now.” They are asking a sharper question: are AI systems actually used in workplaces, or are we just guessing based on theoretical capability? It’s the difference between owning a gym membership and showing up at 6 a.m. every day. Capability exists either way—but impact only happens when you show up. Their solution is a new metric: Observed Exposure. It measures not just whether AI could help with a task, but whether it is helping. They combine three inputs: O*NET task data across 800 occupations Estimates of whether LLMs can theoretically speed up those tasks Real usage data from Claude The key: not all AI use is equal. A marketer brainstorming five headlines with Claude is not the same as a support team automating customer queries at scale. One augments work. The other nudges toward replacement. Observed Exposure gives full weight to automated, workflow-integrated AI, and only partial weight to assistive use. The result: a grounded, realistic view. Jobs aren’t being swept away by AI—they are being nudged, augmented, and tested. Panicking over “AI can touch this job, therefore it’s doomed” misses the bigger picture.