Anthropic Just Proved That AI Literacy Is the New Job Security
There is a question that follows anyone who covers AI. Anyone who teaches it. Anyone who works near it. The question is simple: Will AI take my job?
It is the wrong question. And Anthropic just published the data that proves it.
In a new interview with Fortune, Peter McCrory, Anthropic’s head of economics, broke down what the company’s landmark labor market study actually means for working professionals. The findings challenge nearly every headline you have read about AI and employment. The real divide forming in the workforce is not between humans and machines. It is between people who are building AI fluency and people who are not.
The Exposure Gap Is Real. The Panic Is Not.
Anthropic’s study, Labor Market Impacts of AI: A New Measure and Early Evidence, introduced something the AI conversation desperately needed: actual usage data. Instead of predicting what AI could do to jobs, the researchers measured what AI is doing. They tracked real enterprise usage of Claude across hundreds of occupations and compared two metrics: theoretical exposure (what AI can handle) and observed exposure (what workers are actually using it for).
The gap between those two numbers is enormous. Computer and math occupations have 94% theoretical exposure but only about 30% observed adoption. That gap is not a failure of the technology. It is the opportunity window.
You are probably wondering why the gap is so large if AI is supposedly so capable. McCrory offered a revealing answer: adoption is concentrated in a small set of tasks within a small number of occupations. Three to four out of every ten conversations on Claude’s platform are coding related, yet coding occupations represent just 3% of the workforce. The rest of the economy is barely scratching the surface.
This is not a story about machines replacing humans. This is a story about most humans not yet knowing how to work with machines.
The Skills That Matter Are Not the Ones You Think
Here is where the research gets genuinely important for anyone building a career in 2026. McCrory shared a finding that should reshape how we think about AI training: the sophistication of Claude’s output directly correlates with the expertise of the person using it.
The researchers measured this by analyzing prompts. How many years of formal education would someone need to write the prompt? How many years would they need to understand Claude’s response? The correlation was nearly perfect, across tasks, across countries. If you bring expertise to the conversation, AI amplifies it. If you bring nothing, you get nothing worth having.
Peter McCrory, Head of Economics, Anthropic
This is the opposite of the narrative that AI makes expertise irrelevant. The data shows that AI makes expertise more valuable than ever. The difference is that expertise now includes knowing how to direct AI effectively, evaluate its output, and integrate it into real workflows. That is AI literacy. And it is rapidly becoming the dividing line in the labor market.
The Real Threat Is Not Displacement. It Is Drift.
The study found no systematic increase in unemployment among workers in AI-exposed occupations. That sounds like good news, and it partly is. Turns out the headlines about mass layoffs were overblown.
The concerning signal is subtler. Anthropic found a 14% decline in the job-finding rate for workers aged 22 to 25 in AI-exposed fields since ChatGPT launched. The jobs are not disappearing. The entry points are narrowing. Companies are not firing experienced workers; they are hiring fewer new ones because AI is absorbing the routine tasks that junior roles traditionally handled.
I call this Competence Drift. The floor of what qualifies as entry-level competence is rising. Employers now expect baseline AI fluency the way they once expected spreadsheet proficiency. Workers who cannot demonstrate that fluency are not being replaced. They are simply not being considered.
A 2026 DataCamp study of 500+ enterprise leaders confirms the pattern. Nearly 60% report an AI skills gap, and the gap is not about advanced engineering. It is about foundational literacy: evaluating whether AI output is accurate, knowing when to use AI and when not to, integrating AI into existing workflows. The IDC projects that sustained AI skills shortages could cost the global economy $5.5 trillion by 2026.
Like what you’re reading? Get insights like this delivered daily.
Exposure Is Not Fatal. Inaction Is.
McCrory used a medical metaphor that deserves repeating: exposure to AI is by no means fatal. The real estate manager who uses Claude to automate admin work frees up time for the interpersonal negotiation that no model can replicate. The microbiologist who offloads data analysis to AI can spend more time collecting samples in the field. Exposure, in these cases, reinforces expertise rather than replacing it.
The key insight is that AI reshapes jobs differently depending on the role. Some workers are being upskilled: AI takes over routine tasks and lets them focus on higher-value work. Others are being deskilled: AI covers the most complex parts of their job, leaving them with simpler tasks and less leverage.
Which outcome you get depends entirely on whether you understand the technology well enough to direct it. That is not a prediction. That is what the data already shows.
What This Means for You (Starting Today)
McCrory’s recommendation was simple and direct: start using the tool. It does not have to be Claude. Just use some AI technology and develop a sense for where it excels and where it falls short. In his words, the experimental process itself is rewarding because it broadens what you are able to do.
That recommendation maps directly onto what we teach at Harvest Kernel through the SeedStacking methodology. You do not need to become a prompt engineer overnight. You need to stack one small AI skill per day until fluency becomes second nature. The compound effect of daily practice is what separates the people building career resilience from the people reading about career anxiety.
McCrory also made a point about cognitive endurance that deserves attention. He argued that even as AI handles more implementation work, the act of learning hard things develops transferable mental muscle. The discipline you build from struggling through a complex analysis does not disappear when AI starts assisting with the calculation. It transfers to the next challenge.
That is the deeper argument for AI literacy. It is not just about keeping your current job. It is about developing the adaptability that makes you valuable regardless of which tasks machines eventually absorb.
Your Takeaway
The AI literacy gap is now measurable. Anthropic’s data shows that the difference between theoretical AI capability and actual workplace adoption is not a technology problem. It is a fluency problem. The workers who close that gap first will define what the next era of work looks like. The ones who wait will be defined by it.
Ready to Go Beyond Reading and Start Building AI Fluency?
The gap between knowing about AI and knowing how to use it is exactly what Harvest Kernel’s Learning Community was built to close. Free courses, daily discussions, and a methodology designed to make AI fluency automatic.
