On March 5, 2026, Anthropic published a research report on AI and jobs that every working professional should read. Not because it's alarming. Because it's specific, and it shows where AI is already covering real work tasks, not only where it could.
Anthropic is the company that builds Claude, one of the most widely used AI systems in the world. When they study how AI is actually affecting jobs, they're not working only from forecasts or surveys. They use data on how Claude is actually used and ask: which tasks is AI actually covering, and in which occupations?
The answer should focus your attention — especially if you work in customer service, administration, data-heavy roles, or any knowledge work that involves producing documents, reports, or structured information.
What They Measured (And Why It Matters)
Most AI-and-jobs research asks: what could AI theoretically do? Anthropic asked a harder question: what is AI actually doing, right now, in real professional settings?
They introduced a new metric called Observed Exposure — a score that combines theoretical AI capability with real-world usage data, weighted toward automated (not just assistive) and work-related uses. The result is a measure of what AI is actually covering, not just what it could do.
One of their most important findings: AI is far from reaching its theoretical capability. The gap between what AI could do and what it is doing is still large. The authors expect observed coverage to grow as adoption deepens, which is a reason to prepare early.
"...we hope future findings will more reliably identify economic disruption than post-hoc analyses."
"This framework is most useful when the effects are ambiguous—and could help identify the most vulnerable jobs before displacement is visible."
The Numbers That Should Stop You Cold
Here are the ten most exposed occupations by observed coverage, according to the report (its Figure 3):
| Occupation | Leading Automated Task | Observed Exposure |
|---|---|---|
| Computer Programmers | Write, update, and maintain software programs | 74.5% |
| Customer Service Representatives | Confer with customers to provide info, take orders, handle complaints | 70.1% |
| Data Entry Keyers | Read source documents and enter data into systems | 67.1% |
| Medical Record Specialists | Compile, abstract, and code patient data | 66.7% |
| Market Research Analysts and Marketing Specialists | Prepare reports of findings, illustrating data graphically and translating complex findings into written text | 64.8% |
| Sales Representatives, Wholesale and Manufacturing, Except Technical and Scientific Products | Contact customers to demonstrate products and solicit orders | 62.8% |
| Financial and Investment Analysts | Inform investment decisions by analyzing financial information to forecast business, industry, or economic conditions | 57.2% |
| Software Quality Assurance Analysts and Testers | Modify software to correct errors or improve performance | 51.9% |
| Information Security Analysts | Perform risk assessments and test data processing security | 48.6% |
| Computer User Support Specialists | Answer user inquiries regarding computer software or hardware operation to resolve problems | 46.8% |
Let that sink in for a moment. Customer service representatives rank second in observed AI exposure, at 70.1%. Not theoretical exposure. Observed, in API-based deployments.
The Signal You Can't Ignore: Young Workers
Here's the finding that deserves the most attention — and gets the least press. Anthropic's research found tentative evidence that the job-finding rate of workers aged 22–25 in AI-exposed occupations has fallen about 14% compared with 2022. The authors call the result just barely statistically significant and note that other explanations are possible. Not layoffs. Not pink slips. Possibly slower hiring.
This matters for two reasons. First, if it holds up, it would be an early signal of AI's effect on the labor market. Second, it doesn't show up in unemployment statistics, which is why most people don't feel it yet. Young people who can't get entry-level jobs in their field don't always show as unemployed. They take different jobs, return to school, or drop out of the labor force entirely.
If you are in an AI-exposed field, or if you have a son, daughter, or employee trying to break in: the door may be narrowing. It is a good time to build the skills that change your category.
The research found no systematic increase in unemployment for highly exposed workers so far. The young-worker result is tentative, but it is the kind of early signal the authors say their framework is meant to watch for. The question is whether you use the time.
Who Is Most Exposed? (This Will Surprise You)
One of the most counterintuitive findings in the entire study: workers in the most AI-exposed occupations are more likely to be female, more educated, and higher-paid than those with zero exposure.
The highest-exposure group is 16 percentage points more likely to be female, earns 47% more on average, and is nearly four times as likely to hold a graduate degree compared to unexposed workers (17.4% versus 4.5%).
This shatters the assumption that AI is coming for factory workers and delivery drivers. The most immediate, observed, right-now exposure is concentrated in white-collar, knowledge-based, credentialed work. The professionals who were supposed to be safe are not.
The Future Is Already Showing Up in the Projections
Anthropic matched their exposure data against Bureau of Labor Statistics employment projections through 2034 and found a slight relationship: for every 10-percentage-point increase in AI coverage, projected employment growth is 0.6 percentage points lower.
Exposure shows where change is concentrated. It does not show jobs already lost; the report finds no rise in unemployment among the most exposed workers so far. The question is whether you are prepared as AI coverage grows.
What This Means for You
The Anthropic study describes national patterns. It does not estimate when AI will reach any particular region or employer, so the practical lesson is to understand your own task mix rather than wait for a local signal.
Customer service is one of the occupations the report highlights. If tasks like these are part of your job, the report is a prompt to look closely at which of them AI could cover.
The professionals who thrive in the next wave won't be the ones who avoided disruption. They'll be the ones who saw it coming, measured their specific exposure, and moved into higher-ground roles before the tide changed.
Why IPERA™ Was Built for Exactly This Moment
Anthropic's Observed Exposure metric works at the task and occupation level, which is also how IPERA™ asks about your work. They are not the same thing: the report is national research on how AI is used, and IPERA™ is a self-assessment about your own situation. The report does not evaluate or endorse IPERA™.
What IPERA™ can do that no national study can: ask about your specific tasks, in your specific role. A national figure means little if you don't know where you personally sit.
If your IPERA™ results show high AI Substitution Risk, this research tells you why that score matters. If your results show strong Adaptive Capacity and AI Fluency, this research tells you why those dimensions are your moat. Either way, the research points to the value of acting while you have time.
Anthropic's economists measure AI exposure at the task and occupation level, and their report finds that actual coverage is still far below what is technically possible. Your IPERA™ results are not a quiz score and not a prediction. They are a structured look at your own tasks and skills, informed by research like this.
What You Should Do Today
Reading research is not a plan. Here is what action looks like:
Step 1: Know your number. If you haven't taken IPERA™, you don't know your actual exposure. Job titles are not exposure measures. Task mix is. The assessment takes about 20 to 30 minutes and gives you a specific score across five dimensions.
Step 2: Act on the dimension that matters most for you. If your AI Substitution Risk is high, the play is not to panic — it's to shift your task mix toward oversight, judgment, and relationship work before your employer does it for you. If your AI Fluency is low, that's the first gap to close.
Step 3: Use your time. Change is gradual, which gives you room to prepare, but only if you use it to build something. The national data already shows early signals. Be ready.
There is time to prepare. It is not unlimited, and it is not waiting for you to be ready.