In March 2026 Anthropic published a research paper called Labor market impacts of AI: A new measure and early evidence. Instead of guessing which jobs AI could replace, they measured which tasks people are actually doing with Claude at work, using a metric they call observed exposure.
The headline finding: the gap between the two is huge, which is exactly why the chart looks scarier than reality. Computer and math roles sit at roughly 94% theoretical and only about 33% observed, so the tools exist but hardly anyone has actually put them to work yet. That gap is the interesting bit. It means the fields flagged as most exposed are also the fields where being the person who already uses this well makes you stand out fastest.
Read the report yourself: anthropic.com/research/labor-market-impacts
This is the part people skip, and it is the part that makes everything below personal instead of theoretical. You are going to hand Claude the actual research, not a headline about it, so its read on your job is grounded in the real numbers.
Paste this with the PDF attached. It stops you reading 17 pages of methodology to find the two paragraphs that matter to you.
I have attached Anthropic's research paper "Labor market impacts of AI: A new measure and early evidence". I work as a [your job title] in [your industry], in [country or region].
Read the paper properly and pull out only the parts that touch work like mine. For each finding you surface, quote or closely paraphrase what the paper actually says, tell me which occupational group it falls under, and give me both the theoretical exposure figure and the observed exposure figure where the paper reports them.
Then do three things. First, tell me plainly whether my occupation sits in the high, middle or low exposure band, and say if the paper does not cover my exact role rather than guessing. Second, explain the gap between theoretical and observed exposure for my group in one short paragraph, in plain language. Third, list anything in the paper that contradicts the scary version of this story, because I want the balanced read, not the headline.
If the paper genuinely does not have data on my field, say so directly instead of filling the gap.Every figure below is from the Anthropic paper itself, not a hot take about it.
📗 Could vs is (the reassuring gap): computer and math roles show about 94% theoretical exposure, but only about 33% observed today. Office and admin roles show about 90% theoretical, and the actual footprint so far is small. Source: Labor market impacts of AI, Anthropic, March 2026.
📕 The most exposed occupations (observed):
What they share: most of the work output is text or code, the kind of thing a model can produce directly. Source: Anthropic, March 2026.
🍳 Who is barely touched: about 30% of workers have zero recorded exposure, including cooks, mechanics, bartenders and lifeguards, because their tasks barely show up in AI usage at all. Source: Anthropic, March 2026.
👀 The plot twist: the most exposed group is more educated, earns about 47% more on average, and is 16 percentage points more likely to be female. The roles people chose specifically to feel safe (well-paid knowledge work) are the ones showing up most. Source: Anthropic, March 2026.
Here is the whole point: instead of staring at a chart, map YOUR real tasks against the exposure data. Send this in the same chat, so Claude still has the paper in front of it, along with a quick brain-dump of your weekly tasks.
You are my career strategist. I'm going to give you (1) my real weekly work tasks and (2) context from Anthropic's 2026 labor market report, which measures "observed exposure" (tasks people already do with AI) vs "theoretical exposure" (tasks AI could do in principle). My role: [your job title]. My typical week, task by task with rough hours on each: - [task 1, hours] - [task 2, hours] - [task 3, hours] - [keep going until it's honest] Do this for me: 1. Sort each of my tasks into three buckets: already automatable today (high observed exposure), automatable soon but not yet common, and hard to automate / human-anchored. 2. For each task, name the specific reason it lands in that bucket (is the output text or code? does it need physical presence, live judgment, relationships, or accountability?). 3. Tell me what percentage of my week sits in each bucket. 4. Give me the 3 highest-leverage skills or task shifts that would move me toward the harder-to-automate side of my own role, specific to my job, not generic advice. 5. Suggest 2 ways I could use AI to do the exposed parts of my job faster, so I become the person who runs the tools instead of the person the tools replace. Be honest, not comforting. If something is genuinely exposed, say so, then tell me what to do about it.
A diagnosis you do nothing with is just anxiety with extra steps. This is the prompt that converts the mapping into something you can actually start on Monday, built around the parts of your role the data says are hardest to automate.
Using the task buckets you just built for me, write a 30-day plan that moves me toward the human-anchored side of my own role. Assume I have five focused hours a week and no budget beyond tools I already pay for, and that I still have to do my actual job the whole time.
Structure it as four weeks. For each week give me one theme, two or three concrete actions with a rough time cost on each, and one visible output I can point at afterwards, like a document, a workflow I set up, or a conversation I had. Every action has to connect back to a specific task from my list, so tell me which one it maps to.
Two rules. Do not give me generic advice like "learn AI" or "build your personal brand", and do not put me on a course I will never finish. I want things that change how my week actually runs.
Finish with a short list of what I should deliberately hand over to AI during those 30 days, so I free up the hours the plan needs, and one sentence on how I will know at the end of the month whether it worked.🧭 1. The 12-month watchlist:
Based on my task breakdown above, build me a simple watchlist: which 3 of my tasks should I expect AI to get noticeably better at in the next 12 months, what would be the early warning sign for each, and what should I start learning now so I'm ahead of it rather than behind it?
💬 2. The manager conversation:
Draft me a short, non-alarmist talking point I could bring to my manager: how I want to use AI to take the repetitive, high-exposure parts of my role off my plate so I can spend more time on the judgment-heavy work only I can do. Make me sound proactive, not panicked.
🚪 3. The adjacent-role map:
Given my current skills and the exposure data, show me 3 adjacent roles or specializations that are lower on the observed-exposure list but reachable from where I am now. For each, tell me the 2 or 3 skills I'd need to add and roughly how long that would take.
Observed exposure measures usage, not job loss. A task showing up in AI usage doesn't mean that job disappears, it often means the work changes shape and the person using the tools gets more valuable. Treat this as a map of where to lean in, not a countdown. And Claude's read is only as good as the tasks you feed it, so be honest about your real week.