TL;DR: The replacement worry is legitimate, and the marketing answers in both directions are useless. The “you won’t be replaced, you’ll be enhanced” talking point is comforting and partly wrong. The “everything will be automated” panic is dramatic and also partly wrong.
AI hits different categories of work in different ways, and the structure of your own day decides which category you’re in. What I tell clients about AI starts there: most professional work falls into three categories, and each one calls for a different response.
Neither camp is worth listening to. The people promising you’ll be upgraded and the people promising you’ll be unemployed both get paid for the headline, and the professional losing sleep over it walks away with nothing to act on.
I’d much sooner you look hard at your own week. Some of your hours are going to shrink and some won’t, and you can tell which is which without waiting for a pundit to decide for you.
What is wrong with asking whether AI will replace your work?
From the podcast
Trevor Sumner refuses both of the comfortable answers, on Leaders and Their Stories:
Anybody who tells you it’s not going to put anybody out of business, we’re all just going to be kind of higher level, higher order, because of this, I think is being intentionally techno-optimistic. And anybody who says it’s just going to put everybody out of jobs is being techno-pessimistic. And as always, the answer is usually somewhere in the middle.Which is why the useful question is structural, not directional: what about your work is pattern, and what’s judgment.
Most discussions pose the replacement question poorly, because the framing assumes a binary outcome.
Either your work survives intact or it disappears. Neither outcome describes what happens when a technology changes the productivity of a field. (See how long new technology takes to show up in productivity numbers) The reality is finer-grained. Specific tasks within a job get cheaper to do. Some tasks become trivial and the work that used to fill them gets redistributed. For more, see AI Isn’t a Fad and when to bring a human back into the work.
Some tasks remain expensive and become the part of the job that defines its value. The job that emerges five years later is recognizably the same job done by different hours of human labor on different parts of the underlying activity.
The “you will not be replaced, you will be enhanced” talking point is comforting and partly wrong.Share on X
The undramatic version doesn’t make a good headline, so most coverage runs the binary one. I think that’s a disservice to anybody whose living depends on the answer. A professional asking whether their work will survive deserves a way to check it for themselves, and the check starts with where your hours go.
Watch: Top 20 Jobs AI Will Replace, The Infographics Show
What is the structural test for AI replacement?
Look at your working week and divide it into three categories of activity.
I’d much sooner you look hard at your own week.
Category one is task work that involves writing, summarizing, organizing, researching, or communicating in text. The second category is judgment work that involves making decisions based on expertise, integrating information across many sources, weighing trade-offs, or producing the analysis that drives a decision. The third is human-presence work. It means being a known person in an ongoing relationship, performing live, applying physical skill, and carrying the trust and rapport with a client that took years to build.
Estimate the fraction of your working hours that goes to each category. The fractions tell you most of what you need to know about how AI will affect your work over the next five years. Roughly, category one is what AI affects most. Category two is what AI assists but doesn’t replace. Category three is what AI doesn’t affect at all, regardless of how good the models get.
Estimate the share of your week in each category to see how AI will change your work over the next five years.
What happens in category one
Task work in text is the category AI changes most dramatically.
Hours you used to spend cleaning up an interview transcript become ten minutes. The hour you used to spend drafting a routine memo becomes fifteen minutes. Afternoon work you used to spend summarizing a stack of research becomes an hour. The category doesn’t disappear. It compresses. The professional doing category-one work without AI assistance is competing against the same professional doing the same work with AI assistance in a third of the time.
My own work shows the pattern. I use AI every day as a digital assistant: it turns interview recordings into transcripts, helps me build the summary I send a client after each session, and hands me a first-pass outline to argue with. Then I write the book myself. The hours it saves me are real, and I’d be a fool to give them back, but none of them were the hours a client hires me for.
That compression is real and uneven across professions. A profession that’s mostly category-one work, like routine technical writing, basic editorial production, or formulaic research summarization, will see serious labor compression and probably some displacement. Most knowledge work has significant category-one work alongside serious category-two or category-three work. In those professions the category-one hours compress while the category-two and category-three hours stay roughly the same. The job won’t disappear. It’ll rebalance toward the categories the tools can’t help with.
What happens in category two
Judgment work is where AI is useful but not replacing. The model can help you organize information, surface considerations you might have missed, draft an outline of an analysis, or check your reasoning against a quick alternative framing.
The work itself, the actual judgment, stays with you because the model doesn’t have the context that makes the judgment yours. It doesn’t know your client, your industry, your case, or what this particular situation requires. It can produce competent generic analysis. You produce the analysis that answers the question.
The professional who does serious category-two work finds that AI assistance speeds up the parts of the judgment work that were always mechanical anyway, like background reading and initial draft generation, while leaving the actual judgment exactly where it was. The net effect is that judgment work becomes faster but not cheaper to the customer, because the judgment is what was being paid for, and the judgment is still scarce. Most of medicine, law, executive consulting, and serious investing falls in this category.
I’d tell a lawyer or a consultant the same thing I’d tell a writer. If a client pays you for judgment and you hand the judgment to a model, you’ve stopped earning the fee. Let the tool do the background reading and keep your own hands on the call.
The work is changing in detail but not in essence. A piece on the realistic labor picture covers more of how this is playing out across professions.
What happens in category three
Human-presence work is the category AI doesn’t affect, today or in any near-future scenario worth planning around. The category includes the surgeon’s hands in the operating room, the therapist’s presence with a patient in distress, the teacher in the classroom whose effect on a student is partly about being that specific person in that specific role, the performer on stage, the craftsperson whose work depends on physical skill developed over years, and the salesperson whose value is the relationship with a particular buyer that took a decade to build.
None of those categories is in any real way replaceable by AI, because the value is the specific human being present in the specific situation. The work is the human itself, beyond text generation, analysis, or organization. Professions that are mostly category-three work, regardless of whether they involve high or low formal education, are the safest categories from AI displacement, and the people working in those professions can stop reading replacement-anxiety articles in any practical sense.
How do you map your work onto the three categories?
Take a working week and track your hours by category. Be honest. A lawyer who tells themselves they spend most of their time on judgment, when they spend most of their time on document drafting, has misidentified their category mix. A consultant who tells themselves they spend most of their time on analysis, when they spend most of their time producing decks that summarize others’ analysis, has the same problem.
That map stings for plenty of people. Nobody enjoys finding out that the job they describe at dinner parties is mostly formatting decks and tidying other people’s drafts. I’d still take that sting now over learning the same thing from a client who stops calling.
The good news in an honest map is that the rebalancing can be planned instead of experienced as surprise. If forty percent of your week is category one and you suspect that fraction will compress, you can start now to develop the category-two and category-three sides of your work, where your future hours will need to land. The professional who maps and plans is in a much better position than the professional who refuses to look and finds out about the compression two years late.
My take
Track one working week by the hour and sort every task without flattering yourself. Deck formatting and routine drafting are the work that compresses. Judgment changes in detail and stays yours, as long as you don’t hand it to a model. Work that depends on you being in the room isn’t going anywhere. I’d take the sting of that list now over hearing it from a client who stops calling, and then I’d start building the parts of the job that last.
What does AI replacement look like when it happens?
The replacement story for jobs that do get displaced is rarely dramatic.
The pattern is gradual erosion in fees, then in volume, then in the firm’s willingness to hire the role at all. A professional doing the work continues to find clients for some years, but the clients get smaller, the projects shorter, and the rates lower. By the time the displacement is obvious, the affected professional has often spent three to five years competing against AI-assisted alternatives without realizing it.
This part is cruel, because nobody announces it. No one sends a memo. A client pays a little less each year, the projects get a little smaller, and the professional blames the economy. By the time the pattern has a name, the years they could have spent moving toward judgment work are already gone. (See also what a Yale economist’s AGI model gets wrong about writers)
The real signal is the rate trajectory and the project-size trajectory, both of which start moving before anyone talks about replacement.
For the professional in category one, watching those trajectories matters. A professional in category two finds the trajectories are usually fine, and the change shows up as faster project turnaround instead of collapsing fees. Category-three professionals see trajectories completely uncorrelated with AI, depending on the underlying market for the human-presence work. That moves on its own schedule. Your category tells you which trajectories to watch and which to ignore.
Count your hours this week. If most of your week is text work a machine can do in a third of the time, start moving toward judgment and presence now, because the market won’t wait until you feel ready. If most of it already sits in those categories, quit reading doom articles and get back to the work people pay you for. Either way, I’d hate for anyone to learn which category they’re in from a shrinking invoice.
Structure decides replacement, and augmentation decides who comes out ahead. The Birth of the Augmented Human covers how AI used correctly makes you more capable.
Frequently Asked Questions
- Study finds ChatGPT boosts worker productivity for some writing tasks, MIT News
- Study: AI tools cause a decline in freelance work and income, at least in the short run, Washington University in St. Louis, Olin Business School
