AI Can Do Far More Than We're Using It For. That's Both Reassuring and Alarming.

Anthropic published a study this month that introduces a concept worth understanding: "observed exposure." It measures the gap between what AI is theoretically capable of doing in a given profession and what professionals are actually using it for.

The gap is enormous. And depending on where you sit, that's either the most reassuring or the most alarming finding in AI research this year.

The Numbers

The study, authored by Maxim Massenkoff and Peter McCrory, analyzed real-world usage data from Claude in professional settings. Here's what they found:

Computer programmers have the highest observed exposure — about 75% of their tasks are already being handled with AI assistance. But most professions are nowhere close to that. Actual adoption is a fraction of what the technology can feasibly do.

About 30% of U.S. workers have zero AI exposure. Cooks, mechanics, bartenders, dishwashers — jobs that require physical presence and manual skill. No language model is replacing those roles anytime soon.

The workers most exposed tend to be older, more educated, female, and higher-paid. These are knowledge workers — the people whose work is primarily about processing information, writing, analyzing, and communicating.

The Reassuring Part

Despite years of AI capability growth, the researchers found no systematic increase in unemployment among highly exposed workers since late 2022. The mass displacement that many predicted hasn't materialized. People are still employed. Companies haven't replaced their workforces with AI.

This tracks with what I see in the field. Most organizations are still in early stages of AI deployment. They're experimenting with tools, running pilots, and gradually incorporating AI into existing workflows. The revolution is happening, but it's happening slowly — limited by organizational readiness, data quality, change management, and the simple human reality that new ways of working take time to adopt.

The Alarming Part

Here's where it gets more sobering. While existing workers aren't losing their jobs in large numbers, the study found suggestive evidence that hiring of younger workers has slowed in AI-exposed occupations. The job-finding rate for young workers in these fields has dropped approximately 14% since the post-ChatGPT era began.

The researchers draw a pointed comparison: during the 2007-2009 financial crisis, U.S. unemployment doubled from 5% to 10%. They suggest that a similar-scale disruption to white-collar employment is within the range of possibility — not as a sudden event, but as a gradual tightening of opportunity.

The risk isn't that everyone gets fired on a Tuesday. It's that companies slowly stop hiring for roles that AI can partially handle, that entry-level positions quietly disappear, and that the career ladder gets shorter for the next generation of knowledge workers.

What Leaders Should Be Thinking About

This study matters because it frames the AI employment question correctly. It's not "will AI take my job?" — a question that invites panic or dismissal depending on your temperament. It's "how do we navigate a transition where AI capability is growing faster than AI adoption, but adoption will inevitably catch up?"

A few things that matter for leaders right now:

Invest in your people, not just your tools. The organizations that deploy AI well don't replace workers — they redeploy them. The question isn't whether to automate routine tasks. It's what you want your people doing instead, and whether you're preparing them for that shift.

Pay attention to the hiring pipeline. If you're quietly hiring fewer junior roles because AI handles the work those roles used to do, you're solving a short-term cost problem while creating a long-term talent problem. Someone still needs to develop the judgment, domain expertise, and institutional knowledge that makes senior workers valuable. That development has to happen somewhere.

Close the capability-adoption gap intentionally. The gap between what AI can do and what your organization is using it for is a strategic asset — but only if you're closing it on purpose, with a plan. Closing it by accident, through reactive tool adoption, creates chaos.

Think in terms of workflows, not headcount. The most productive frame isn't "how many people can AI replace?" It's "which parts of which workflows should AI handle, and what does that free our people to do?" That question leads to better decisions, better morale, and better results.

The Bottom Line

AI is capable of far more than most organizations are asking it to do. That gap won't last forever. The companies and leaders who use this window to thoughtfully redesign how work gets done — investing in both technology and people — will be in a fundamentally different position than those who waited to see what happens.

Waiting is a strategy. It's just not a good one.