It seems like just about everyone right now is trying to predict which jobs are at risk from AI, and at what level of urgency. Over the past couple of weeks a range of studies and data sets have come out attempting to help paint a clearer picture.

For me, everything remains quite confusing, but a few points felt worth sharing.

The more your job lives on a screen, the more exposed it is. Anthropic's research found that computer programmers have 75% of their tasks covered by AI, customer service reps 65%, and data entry workers 67%. Meanwhile, 30% of the workforce (cooks, bartenders, mechanics, lifeguards) has zero AI coverage.

But exposure alone isn’t the full picture. A new study from GovAI and Brookings mapped occupations on two axes: how exposed a job is to AI, and how easily the worker could pivot to other good-paying work. Web designers and secretaries both score high on exposure. But web designers have transferable skills and education that give them options. Secretaries and clerical workers land in a trickier zone: highly exposed with the fewest alternatives. What separates the two groups isn't the job title. It's breadth of experience, education, savings, and age.

Women are disproportionately in the most vulnerable category. The GovAI/Brookings research found that women hold 86% of those high-exposure, low-adaptability roles. Anthropic's data shows this from a different angle: workers most exposed to AI are 16 percentage points more likely to be female.

My thoughts: It is super frustrating to me that AI could possibly present yet another disproportionate hurdle for women in the workforce. 🤬 But this aside, rather than trying to pick "AI-proof" paths for our kids, we're probably better off building what separates adaptable workers from vulnerable ones: curiosity, breadth, and the ability to learn something new.

I attended a meeting of our school district's AI task force this week. The big topic: whether to allow AI tools for middle schoolers.

None of the group were AI experts. They were educators, administrators, and parents who know academics deeply, being asked to make a consequential decision with no real foresight into how it plays out. And every district in the country seems to be figuring this out alone. Yes, they talk and share with one another, but there's no gold standard or top-down guidance to follow.

Many districts, ours included, are leaning toward turning on Gemini and NotebookLM for middle schoolers, as part of the Google for Education suite. This concerns me… a 6th grader is not a high schooler. Younger kids are far more susceptible to cognitive offloading, unhealthy AI relationships, and falling for hallucinations and deepfakes. Common Sense Media graded Gemini for K-12 as "High Risk." Moreover, there's almost no budget or resources for teacher training or AI literacy courses for students.

It made clear to me that our kids are going to need more support from home on this one. Not because schools aren't trying, but because the pace of AI is outrunning what any district can possibly keep up with. Our kids will have access. The conversations about how to use it well need to come with it.

Speaking of AI in the context of education, there are a whole host of ways you can use it as a personalized learning tutor, from Google’s NotebookLM, guided learning features in Gemini and ChatGPT, and specialized edu AI tools like Khanmigo. A new one I discovered this week (although it has been there for awhile!) is learning mode in Claude.

Whether you’re using Claude Cowork or just the regular Claude chatbot, you can toggle on “Learning” through the tool selector and ask it to help you learn about any possible topic.

I asked it to help me and my son learn about the gold rush and after a few clarifying questions (which I appreciated!) I had a full age-appropriate learning guide with stories, cool insights, interesting charts and graphs, activity sheets, and more.

Moreover, this week Anthropic launched the ability for Claude to generate interactive charts and diagrams directly in the chat. It will automatically determine if/when this would be helpful but also responds to prompts like “make an interactive chart of the periodic table of elements” or “make me an interactive timeline of the gold rush.”

Give these features a try and let me know what you think!

This story has made the rounds, but is an inspiring example of the empowerment AI tools give to all of us. A Sydney AI consultant named Paul Conygham was told his dog had months to live after being diagnosed with a terminal mast cell cancer.

He used ChatGPT, Google’s AlphaFold, and Grok to map research and model Rosie’s mutations - and then partnered with UNSW’s genomics lab to generate a custom mRNA vaccine - that has reportedly shrunk the tumor by ~75%.