In partnership with
Hey {{first_name | there}},
So, Anthropic just revealed the 10 jobs at highest risk of being replaced by AI.
And this isn't the usual fear-mongering "AI is coming for your job" stuff people have been saying for years. This report is based on actual Claude usage data.
They looked at millions of real conversations, crossed them with US labor data, and built a map of which jobs AI is genuinely being used for right now. Not theoretically. Today.
Here are those jobs. The percentage = how much of that job's daily tasks AI is already handling.
Computer Programmers — 75%
Customer Service Reps — 70%
Data Entry Keyers — 67%
Medical Records Specialists — 67%
Market Research Analysts — 65%
Sales Representatives — 63%
Financial Analysts — 57%
Software QA Analysts — 52%
Information Security Analysts — 49%
Computer Support Specialists — 47%
Two things worth noting. First, Anthropic says this is still nowhere near AI's theoretical ceiling; the actual exposure is going to keep climbing.
Second, hiring of young workers (22-25) into these roles has already started slowing down. Unemployment hasn't moved yet, but hiring is the leading indicator.
So what should you do if your job is on this list? Start using AI tools in your actual workflow now, not to experiment, but to get good at directing them. The people who keep these jobs will be the ones who know how to get the best out of AI, catch its mistakes, and handle the parts it gets wrong. That skill is what you're building time for right now.
Which jobs are completely safe? Anything that requires you to physically be somewhere. Cooks, bartenders, mechanics, lifeguards, electricians. AI coverage there is basically zero. If your hands are part of the job, you're fine for the foreseeable future.
What about everyone else? Most jobs aren't fully on the list or fully off it. AI is taking over specific tasks within jobs, not whole jobs overnight. The risk is for people who only do the tasks AI is good at, and nothing else.
🛠️ AI Tools Worth Checking Out
Rocket: Generate and launch landing pages with 25,000+ AI templates you can customize and publish in minutes.
NextDocs: An AI-native editor for docs and slides with built-in LaTeX equations, structured tables, and syntax-highlighted code blocks.
Figr AI: An AI UX design assistant that understands product logic before generating UI designs.
Idea2Clip: Turn a simple idea into a complete AI-generated video with scripts, scenes, and consistent characters.
SurfSense: An AI research and knowledge management assistant that organizes information into searchable insights.
Other AI News You Should Know
OpenAI shipped GPT-5.4 two days after 5.3. It scored 75% on a benchmark for real desktop navigation, the human baseline is 72.4%. It also beat professionals 83% of the time on a knowledge-work test across 44 jobs, and supports 1M token context for long multi-step tasks. Available now for Plus, Team, and Pro users.
The significance here isn't just the benchmarks. It's that a general-purpose AI just crossed human performance on actually using a computer. That's the unlock for agentic workflows most people have been waiting for.
Claude Code just got scheduled tasks. You set it up once, it runs on its own clock. Overnight commit reviews, dependency audits, PR checks, no prompting, no babysitting.
This sounds like a small update but it's actually a meaningful shift. A coding assistant that waits for your prompt is a tool. One that operates on its own schedule is closer to an autonomous agent. Developers are already sharing demos of fully automated workflows running hands-off.
Google quietly released an open-source command-line tool called gws that gives you (and AI agents) full access to Drive, Gmail, Calendar, Sheets, Docs, and more from the terminal. It reads Google's own API directory at runtime so it updates automatically when Google adds new features. Already at 15k stars on GitHub in under a week.
The "AI eating itself" problem is real
A thread this week on AI models training on AI-generated content went viral. The concern: models trained on synthetic output get progressively worse at sounding human. Researchers have been flagging this for a while. Next time something reads as technically correct but weirdly hollow, that's probably why.
This matters because a lot of what you read online now, blogs, summaries, social posts, is already synthetic. If models keep training on that output, the quality of AI responses degrades over time. It's not happening overnight but researchers say the signal is already there. Something to keep in mind when you're using AI output as a source for anything important.
Something To Think About
The Anthropic report and GPT-5.4 landing in the same week isn't a coincidence, it's just the pace now. A new frontier model every few days, and data showing real job exposure moving in the background.
If your job is on that list, the question isn't whether to worry. It's whether you're the person using these tools or the person being replaced by someone who is.
That's all for this week.
- Aashish
What did you think about today's newsletter?
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.