"Will AI Make Me Obsolete?" Stop. Ask This Instead

I’ve been in public health for over 25 years. (I’m flabbergaster every time I do that math. It seems like just yesterday that I was living in Ann Arbor, Michigan, studying health behavior change, and eating all the Zingerman’s reuben sandwiches my student budget could afford.)

Since then, I’ve written hundred of grant proposals, conducted countless needs assessments, sat through more task force meetings than I care to remember, and tried to explain p-values to people who’d rather be literally anywhere else.

So when the AI conversation turned to the subject of job impact, something bugged me. Most of it was either terrifying (”AI will replace you!”) or breezy and unhelpful (”AI is just a tool!”). Almost none of it was specific.

Meanwhile, the same question keeps showing up in my inbox, in DMs, and in the awkward post-presentation Q&A:

How is AI going to affect my job?

Honestly? I have struggled to answer.

I can tell you, in great detail, how AI has reshaped my work. Writing. Gathering information. Teaching. Public speaking. Analyzing data. Monitoring policy and news. Those are the tasks that fill my days, so those are the tasks I have data on.

But your day might look completely different. You might be teaching diabetes self-management in a church basement. You might be inspecting restaurants for code violations. You might be standing in waders pulling water samples from a polluted creek.

I cannot answer “how will AI affect your job?” because I don’t do your job.

So I built something that I think will help.

A county health department epidemiologist, a community health worker doing door-to-door outreach in a rural county, a environmental health researcher at a university, and a public health nurse coordinating care for high-risk patients all carry some version of the title “public health professional.” They are not doing the same job. Not even close.

Tools do exist for assessing AI exposure by job. A few are even pretty well-designed. Andrej Karpathy (a well-known AI researcher) built a dashboard scoring more than 300 occupations by title. Sites like AIExposure.org cover 900-plus job categories using federal labor data. Academic economists have published serious work modeling unemployment risk by occupation.

I’ve poked at all of them. When I search for something like “health educator” or “epidemiologist,” I get a number. A score. A percentile. What I don’t get is anything that reflects how different the work of a disease surveillance epidemiologist is from a community health program evaluator, even if they share a job family. These tools were built on Bureau of Labor Statistics occupation codes, which are themselves very blunt instruments. They were not built with public health in mind, and it shows.

Here’s the thing: AI doesn’t affect jobs. It affects tasks. And every public health job is a bundle of tasks, some wildly different from the others. Some involve sitting in front of data on a screen. Some involve standing on a doorstep trying to earn someone’s trust. Those are not the same kind of work, and they are not going to be affected by AI in the same way or on the same timeline.

That’s the gap. We keep talking about AI and “the workforce” when we should be talking about AI and the work.

I didn’t set out to publish an app. This project started, at least partly, as an excuse to play around with a new tool.

The platform is called Lovable. Lovable.dev, which, importantly, is not lovable.com. I know this because when I tried to navigate to lovable.com on my home wifi, my browser threw up a blocked-content screen. Turns out I have kid filters on my router. I did not know I had kid filters on my router. My kid, who was standing nearby, calmly identified the blocked screen, explained what it meant, and probably also knows seventeen ways to get around it. (We will be having that conversation soon.)

Lovable is an AI-assisted development platform that lets people who are not developers, like me, build functional apps. You describe what you want in plain language, and the AI writes the actual code (Python, HTML, TypeScript, all the languages with semicolons that strike fear in the hearts of normal people). It feels like witchcraft. The modern-day term for it is vibe coding.

But here’s the thing about playing around: sometimes you end up building something real. About ten to twelve hours of scattered evenings and weekends later, what started as curiosity about a new tool turned into a genuine attempt to answer the question I couldn’t shake. Does anything like this exist for public health workers specifically? The answer, as it turns out, is no. And that felt like a reason to keep going.

The whole assessment hinges on three categories, so let’s get on the same page.

Automatable tasks are things AI can do on its own, or nearly so, with a human mainly checking the output.

The classic non-AI example? The Scantron. (If you are too young to remember Scantrons, congratulations on your youth, and please tell me what your joints feel like.) A student bubbled in their answers with a No. 2 pencil. A machine read those bubbles, compared them to an answer key, and spit out a grade. A human still passed out the forms and troubleshot the occasional jam, but the scoring itself? Done by the machine.

In public health, automatable tasks include routine report generation, certain kinds of data cleaning, and news monitoring. If you spend a significant chunk of your week here, that part of your work is likely to change.

Augmented tasks are things where AI makes you substantially more effective, but where you still need to be in the driver’s seat.

Take grant writing. I give an RFP (Request for Proposals, the document that explains what a funder will pay for) to a chatbot. I ask it to summarize the priorities. I share past proposals and ask it to suggest what to reuse. I draft a section, then ask the chatbot to critique it. I am still doing the strategic thinking. The chatbot is doing the heavy lifting on synthesis, reformatting, and second-opinion offering. AI is an incredible partner here. But it’s a partner, not a replacement.

What I would not want to do is fully automate grant writing, as in: my computer scrapes the internet for funding opportunities, picks ones that fit, drafts proposals from my old materials, and submits them without my review. That level of automation is somewhere between a bad idea and a nightmare. Strategy belongs to humans.

Deeply human tasks are the ones where the value is intrinsic to the human doing them.

Showing up for a community that has reason to distrust institutions. Making an ethical call in a gray area. Being accountable to a regulatory body. Holding space for someone who just got difficult news. Navigating a complicated political environment where relationships are everything. These tasks aren’t just “safe from AI” because the technology isn’t ready. They’re different in kind. They depend on things that can’t be processed: trust, accountability, physical presence, lived experience, moral reasoning.

This framing changes the question from “will AI replace me?” to “which parts of my work are going to change, and how do I want to position myself?” Those are more useful questions. They’re also less anxiety-inducing, which in 2026 feels like a public service.

For those curious how the sausage was made:

First, I identified the most common public health jobs and the tasks that make up those jobs on average. I pulled from a pile of sources: U.S. Department of Labor occupational data, PH WINS (the Public Health Workforce Interests and Needs Survey), and a handful of occupational guides and association reports.

The first version of the task list was too broad. The categories matched how we usually describe public health work at the 30,000-foot level. Technically accurate, practically useless. So I revised. Hard. I thought about what people really do on a Tuesday morning. That process led to a much more specific list of work areas, including:

• Surveillance and disease monitoring (tracking patterns with structured data)

• Case investigation (interviewing someone who may have been exposed to something, navigating trust, fear, and sometimes language barriers)

• Health equity analysis (asking not just what is happening, but who it’s happening to and why, which requires contextual and structural thinking AI is genuinely not great at yet)

• Community engagement and outreach (relationship-based, presence-based, trust-based, and not really replicable by a chatbot, no matter what the hype says)

• Inspection and compliance (showing up, observing, making judgment calls in real time, and sometimes having uncomfortable conversations with people who really don’t want you there)

• Informatics and data systems (more exposed to AI than most others, because so much of it involves structured, processable information)

Then, for each task, I assessed the likelihood it would be automated and/or augmented by AI, using the framework above.

The model is intentionally task-based, not title-based, because two people with the same job title can have wildly different day-to-day realities. A “health educator” at a federally qualified health center is doing very different work from a “health educator” at a state health department, and the AI implications are different too.

Is the model perfect? Nope. Is it more useful than the generic exposure dashboards floating around the internet? I think so. But you tell me.

The app is in alpha, which is a polite way of saying it works, mostly, and I want to know where it doesn’t. The results are estimates, not pronouncements. It is not going to tell you whether to leave public health or predict your career five years out. It is meant to help you think. That’s it.

One deliberate choice I want to name: you can take the assessment without creating an account first. Because asking someone to hand over their email before they’ve seen a single thing of value is exactly the friction that makes people click away. We can do better than that.

You are the very first people I’m sharing this with. I’ll roll it out more broadly (probably starting on LinkedIn) at some later date. But you are the soft launch.

👉 Try the assessment here

Then tell me:

• Did the questions reflect the work you actually do?

• Were the results useful? Surprising? Oversimplified? Dead wrong? Kind of brilliant?

• What is missing?

• What would make this more useful for your team or your students?

Leave a comment, hit reply, or DM me. Every piece of feedback shapes the next version.

If any of this has you curious about building your own thing: do it. Seriously. Lovable has made it genuinely possible for non-developers to build functional tools for public health problems nobody else is going to build for us. If you’re sitting on a workflow nightmare or a question that keeps coming up in your work, the barrier is lower than it has ever been. I want to see what you make.

Two questions I’m sitting with, and I’d love your answers:

Which part of your work do you most want AI to stay away from?

Is there a tool you wish existed for public health that nobody has built yet?

Send me a message or leave a comment. I’m listening.

Leave a comment

P.S. AI helped me draft this newsletter. ChatGPT kept trying to sneak in the word "transformative." Claude kept offering to add bullet points I didn't need. I overruled them both, mostly. The ideas are mine. The em-dash-free sentences took real discipline.