75% of Programming Tasks Are Already Done by AI. Here’s What That Means for Your Career

The Window Is Open — But It’s Closing Fast

Here’s a number that should stop every professional mid-scroll: 74.5%. That’s the percentage of programming tasks that AI is already performing in real workplaces — not theoretically, not in lab demonstrations, but right now, in actual professional settings. Customer service? 70.1%. Data entry? 67.1%.

These numbers come from Anthropic’s March 2026 “Labor Market Impacts of AI” report, and they introduce something the AI conversation has been missing: a measure of what AI is actually doing versus what it could do. The researchers call it “observed exposure,” and the gap between those two numbers is simultaneously the most reassuring and most alarming data point in the entire AI-and-jobs debate.

Now, you might be thinking: “Another AI doom report? I’ve been hearing this for three years.” Fair. But this one is different — and the difference matters. Because nested inside the scary numbers is a roadmap. The same data that shows which jobs are most exposed also reveals exactly which skills make workers more adaptable, not less. The question isn’t whether AI will change your work. It’s whether you’ll be the person using AI or the person replaced by it.

What “Observed Exposure” Actually Means — And Why It Changes the Conversation

For three years, the AI-and-jobs debate has been stuck between two unhelpful extremes: “AI will replace everyone” and “AI is just another productivity tool.” Anthropic’s researchers — economists Maxim Massenkoff and Peter McCrory — cut through both by asking a more grounded question: What work is AI actually being used for in professional settings right now?

Their answer combines three data sources: the U.S. Department of Labor’s O*NET occupational task database (which catalogs tasks across 800+ occupations), theoretical exposure estimates from foundational research by Eloundou et al. published in Science, and — critically — real-world usage data from millions of Claude conversations in professional contexts.¹

The result is striking. In computer and math occupations, AI could theoretically handle 94.3% of tasks. But actual usage covers just 33%. That gap — the distance between blue (what’s possible) and red (what’s happening) — is your window of opportunity. But here’s the part that demands attention: that window is narrowing. As Anthropic’s researchers note, “As capabilities advance, adoption spreads, and deployment deepens, the red area will grow to cover the blue.”²

Translation? The 61-point gap between theoretical and observed isn’t permanent safety. It’s a countdown.

Building AI literacy is no longer optional — it’s career insurance. Start building yours today

Who’s Actually at Risk — And It’s Not Who You Think

If you assumed AI displacement would hit the same workers automation has always hit — factory floors, manual labor, blue-collar trades — this data flips that assumption completely. The workers most exposed to AI aren’t low-wage or low-skill. They’re higher-paid, college-educated, and disproportionately female.

According to the Anthropic report, workers in the most AI-exposed occupations earn 47% more on average ($32.69/hour versus $22.23/hour) and are substantially more likely to hold graduate degrees — 17.4% compared to just 4.5% in zero-exposure roles. Meanwhile, a companion analysis from GovAI and the Brookings Institution, highlighted by The Washington Post, found that women make up approximately 86% of the most vulnerable worker population.³

This isn’t the automation wave of the past. This one is coming for the knowledge economy first — marketing analysts, financial planners, administrative professionals, customer service teams. The very workers who were told their education and experience made them irreplaceable.

The Entry-Level Warning Signal Everyone Should Watch

Here’s the data point that should concern every parent, every career counselor, and every hiring manager: a 14% decline in hiring of workers aged 22 to 25 in AI-exposed occupations since late 2022.⁵

The unemployment rate for highly exposed workers hasn’t spiked — Anthropic’s researchers are careful to note that. But the hiring pipeline is constricting at the entry point. Young professionals entering the workforce are finding fewer on-ramps into the very careers they trained for. It’s not mass layoffs. It’s a quiet narrowing of opportunity that won’t show up in unemployment statistics for years.

If you’re thinking, “That’s just for tech workers” — think again. The same pattern is emerging across business, finance, marketing, and administrative roles. Every occupation group with more than 50% theoretical AI capability is seeing some version of this entry-level squeeze.

Why This Isn’t a Doomsday Story — It’s a Strategy Signal

But here’s what the headlines miss — and it’s the most important finding in the entire report. For almost every job, AI augmentation (helping humans work better) far exceeds automation (replacing humans entirely). The researchers found that augmentative use still dominates how professionals interact with AI tools in practice.⁶

That distinction matters enormously. Augmentation means AI is making existing workers more capable, not fewer. It means the professionals who learn to work with AI aren’t losing ground — they’re gaining it. The threat isn’t AI itself. It’s the gap between workers who understand how to leverage these tools and those who don’t.

Fortune magazine described the potential as a “Great Recession for white-collar workers” — and that framing is useful, because recessions don’t destroy all jobs. They redistribute opportunity. The workers who adapt thrive. The ones who freeze get left behind.

Workers will not lose jobs directly to AI systems but to people who use AI tools more effectively.

Industry analysis, March 2026

The SeedStacking Response: Small Skills, Compounding Returns

So what do you actually do with this information?

The answer isn’t panic-enrolling in a coding bootcamp or pretending AI doesn’t apply to your field. It’s building what we call AI fluency through SeedStacking — the practice of identifying small, high-leverage AI skills and stacking them daily until they compound into genuine capability.

Start with your own “observed exposure.” Look at the tasks that fill your workday. Which ones could an AI tool handle faster? Those aren’t threats — they’re the first skills to stack. Learn to prompt effectively. Learn to evaluate AI output critically. Learn to integrate AI into your existing workflow instead of bolting it on as an afterthought.

The professionals who will navigate this transition aren’t the most technical. They’re the most AI-literate — meaning they understand not just how to use the tools, but when to trust them, when to question them, and when to override them entirely. That’s the difference between someone who gets augmented and someone who gets automated.

Here’s the thing most people miss about AI anxiety: the antidote isn’t more information. It’s more practice. Every day you spend not building AI fluency, that observed exposure gap narrows a little more — and your window of adaptation shrinks with it.

Your career’s AI exposure isn’t going to decrease. Start building AI fluency now — it’s free

The Bottom Line: Exposure Isn’t Destiny — But Inaction Is

Anthropic’s report proves something we’ve been saying at Harvest Kernel since day one: the divide isn’t between people who understand AI and people who don’t. It’s between people who are building AI fluency and people who are waiting to see what happens.

The data is clear. The theoretical capability is massive. The actual adoption is still early. That gap is your opportunity. But gaps close — and this one is closing faster than most people realize.

Don’t wait for your job description to change. Don’t wait for your company to offer AI training. Don’t wait for the unemployment data to catch up to the hiring data. The signal is already here. The question is whether you’ll act on it while the window is still open.

The professionals who start today will be the ones still standing tomorrow. Join the Harvest Kernel community and start stacking

Sources

¹ Massenkoff, M. & McCrory, P. (2026). “Labor Market Impacts of AI: A New Measure and Early Evidence.” Anthropic Research. anthropic.com/research/labor-market-impacts
² Anthropic Economic Index, January 2026. anthropic.com/economic-index
³ Metz, C. & Merrill, J.B. (2026). “See which jobs are most threatened by AI.” The Washington Post. washingtonpost.com
⁴ Yanatma, S. (2026). “How AI will reshape work: Anthropic identifies the most exposed jobs.” Euronews. euronews.com
⁵ CBS News (2026). “Anthropic is tracking which jobs are most exposed to AI.” cbsnews.com
⁶ Anthropic Economic Index Report, January 2026. anthropic.com