The first jobs AI takes may belong to its power users

Good evening. The Owl here.

The people who talk about AI are mostly people who sit close to it.

Almost every day they try a new model, test prompts, and marvel at the text, code and images it produces. And precisely because they know how good it is, they cry out: “This will take our jobs.” “There will be nothing left for humans to do.”

Those voices matter, of course.

Lately, though, I’ve started to see it a little differently.

Maybe the people sounding the loudest alarm about AI aren’t seeing the future before anyone else. Maybe they’re standing in a spot where the sand under their feet has already started to give way.

Picture someone at the rim of an antlion pit (the sand funnel an antlion digs so that ants slide down into it), shouting outward, “It’s dangerous here!” They believe they’re issuing a warning. But their voice is also the sound of the place that is slowly pulling them in.

ChatGPT was released on November 30, 2022. More than three years have passed since then.[1]

So has AI already wiped out every kind of job?

Not so far, at least. Stanford HAI’s AI Index 2026 concludes that large-scale job losses from AI do not yet show up in aggregate employment data.[2]

That doesn’t mean there’s no effect at all. Where the effects do appear, they are actually quite clear.

So where is the ground giving way first?

In this piece I want to think through a paradox: the first jobs AI takes may not be the ones far from AI, but the ones closest to it.

---

AI job loss is not yet “total war”

“AI will wipe out jobs” is a very strong claim.

But the stronger the claim, the more it pays to narrow down where it actually applies.

In the US, for example, Challenger, Gray & Christmas’s June 2026 report counted 101,743 job cuts year to date that cited AI as the reason, about 23% of the total. Since the firm began tracking AI as a separate reason in 2023, the figure comes to 173,568.[3]

That is not a number to brush aside.

Still, if you look at the US labor market as a whole, AI is not replacing every occupation in one sweep. Goldman Sachs explains that labor displacement from AI is partly offset by areas where AI complements human work and adds jobs.[4]

In other words, what’s happening now is not “AI takes everyone’s job at the same speed.”

It’s far more lopsided than that.

Stanford HAI’s AI Index 2026 sums it up this way: AI’s effect on the labor market is concentrated on young workers and on the hiring pipelines of occupations most exposed to AI. Employment of software developers aged 22 to 25, in particular, is said to have fallen by about 20% since 2024.[2]

That is a very important clue.

AI is not suddenly swallowing whole factories, stores or hospitals from the outside edges of society.

What it swallows first is work that is digital, already put into words, and easy to finish entirely on a screen.

Writing text.

Writing code.

Making slide decks.

Summarizing research.

Answering inquiries.

Translating.

Writing ad copy.

Cleaning up meeting minutes.

In short, the work of the people who use AI the most.

What AI does best sits close to “jobs that use AI”

What makes this tricky is that the more someone knows about AI, the closer their work tends to sit to what AI does best.

AI consultants.

Prompt-writing instructors.

People who help companies streamline operations.

Article writing, deck building, code generation and research-for-hire done with AI tools.

These jobs have value, of course. When a new tool arrives, you need interpreters and guides. You need people who set internal rules, people who fit the tool into actual workflows, and people who prevent mistakes.

But there is a question of shelf life.

The job of teaching people how to use AI thins out as AI itself starts teaching people how to use it.

The job of crafting clever prompts to produce text thins out as models get smarter, as tools carry their own context, and as routine tasks shrink to a single click.

Making minutes with AI, summarizing with AI, drafting a first-pass slide deck with AI: that kind of work sits exactly where AI is easiest to automate.

The irony is that the closer you get to AI, the stronger you look at first.

You can say, “I can use AI.”

You can say, “I can do it faster than others.”

You can say, “I can transform your operations with AI.”

But if your only strength is “I can use AI,” the next model update may knock your price down overnight.

It’s natural that people who work with AI are amazed by it and warn about its risks. Their own jobs are the ones sitting nearest to AI’s capability curve.

To go back to the antlion pit: the person treading sand at the rim hears the ground crumbling far more keenly than someone watching the pit from a distance.

How this differs from the dot-com boom

This is where the usual comparison comes up.

Is AI the biggest revolution since the internet?

On investment alone, the AI boom is certainly enormous. According to the OECD, venture investment in AI companies worldwide reached $258.7 billion in 2025, 61% of all VC investment. That share has doubled from 30% in 2022.[5]

Stanford HAI also reports that private AI investment in the US reached $285.9 billion in 2025.[6]

A staggering amount of money is pouring in.

But the way that money flows is somewhat different from the dot-com era.

In the early internet, the front doors of existing industries were rebuilt one after another: selling books online, booking travel online, buying stocks online, advertising online. There was a bubble, of course, and plenty of sloppy businesses. Even so, internet companies and services sprang up everywhere, and a whole wave of companies changed how ordinary people actually lived.

New companies are being born in AI too. Corporate spending on generative AI is rising. Menlo Ventures estimates enterprise spending on generative AI at $37 billion in 2025, 3.2 times the previous year’s $11.5 billion.[7]

Even so, the heat is concentrated in a few places.

Foundation models.

AI infrastructure.

Cloud.

Semiconductors.

Large-scale training and inference environments.

And the giant companies close to all of those.

In other words, at the center of the AI boom, the dominant pattern is not the dot-com-bubble sense that “anyone can start a new internet company.” It’s capital concentrating in a handful of giant models and giant infrastructure.

That doesn’t mean AI is small. Quite the opposite. It’s precisely because it’s so big that it’s hard for ordinary companies or individuals to compete head-on in the foundation-model race.

So outside that race, the excitement can look surprisingly cool.

That may be especially true in Japan. According to the Ministry of Internal Affairs and Communications’ (MIC) 2025 White Paper on Information and Communications, the share of individuals in Japan who had used generative AI rose to 26.7% in the survey for fiscal 2024 (April 2024–March 2025), but that is still low compared with 68.8% in the US, 59.2% in Germany and 81.2% in China. Only about 50% of Japanese companies have drawn up a policy on using generative AI, which also tends to lag other countries.[8]

AI news flows in every day.

But do you see AI startup signs multiplying around town, or an explosion of new AI services in every industry? It doesn’t feel that way.

Many of the people who are excited are those who sell AI itself, help companies adopt AI, publish content about AI, or make work more efficient with AI.

Which is why I wonder whether what AI is changing first is not society as a whole, but the work of “the people who talk about AI, use AI, and make their living with AI.”

Seen that way, the odd feeling I’ve had lately starts to make a bit more sense.

The shelf life of AI consulting

The phrase “AI consultant” has a slightly strange ring to it.

It sounds both necessary and a little dubious.

In practice, a company that wants to adopt AI needs to sort out its workflows, clean up its data, manage access rights, handle security, train staff and set up metrics to measure results. Signing a contract for a tool is not the end of it.

McKinsey’s 2025 survey also finds that organizations getting value from AI have management practices in place, such as processes that define when humans must validate model outputs.[9]

That part is real work.

AI consulting gets shaky, however, when all it amounts to is “teaching people how to use AI,” “introducing handy tools” or “handing over a collection of prompts.”

Because that part keeps getting commoditized.

“I can use Excel” used to be a strength. But it’s hard to stay highly valued for long on basic Excel skills alone. The value stayed with people who understood accounting, inventory management or sales management. In other words, people who grasped what the work meant, not Excel.

AI is the same.

Being able to use AI is the entry point.

But that entry point keeps getting wider, shallower and cheaper.

What remains is what lies beyond it.

Which work should change?

Which decisions stay with humans?

Which failures are acceptable, and which must be avoided at all costs?

How do you measure results?

Did the experience of customers and employees actually get better?

AI work that can’t take hold of these questions is the work AI can most easily replace.

It’s work that sits close to AI yet holds none of the responsibility that lies outside it.

Are “people who master AI” safe?

There’s a line you hear a lot:

“AI won’t take your job. Someone who uses AI will.”

I think that’s half right.

People who use AI are faster than people who don’t. That’s happening in many jobs: research, drafting, translation, summaries, coding help, preparing for customer support. The gap between those who use it and those who don’t is already quite large.

But there’s another half.

AI takes the jobs of people who use AI too.

More precisely, it starts with jobs whose value is “using AI” itself.

That’s a pretty unpleasant thought.

The people who jump on a new tool early benefit first. That much is natural. But as the tool matures, the early adopters’ know-how gets absorbed into the tool.

Prompting tricks get absorbed into the UI.

Summary formats get absorbed into templates.

Code-completion skills get absorbed into the development environment.

The steps for building a deck get absorbed into agents.

Once that happens, “I can use AI” no longer sets you apart.

If anything, the closer a task is to AI, the more its price drops every time AI improves.

So the people who are truly safe are not “the people who have mastered AI.”

They are the people who use AI and also hold what AI cannot.

Context about the customer.

A feel for the industry.

Responsible judgment.

The ability to get things moving inside an organization.

The resolve to take the blame when things go wrong.

The ability to see things through to results, not just deliverables.

When AI takes jobs, it first goes after work that is “intellectual, digital, and produces easily standardized output.”

The people doing that work know best how impressive AI is.

So they cry out.

And their cry is largely right.

But it is not only a prophecy about society as a whole. It is also a report on the ground shifting under their own feet.

So where do you go?

After all this, it might sound as if using AI is dangerous.

But my conclusion is the opposite.

This is not about staying safe by keeping away from AI.

It’s about the danger of merely getting close to AI.

What matters is not working near AI, but taking on the responsibilities that sit outside it.

If you write, it’s not about producing text quickly with AI. It’s about judging what should be said, what should be left unsaid, and how it should reach which readers.

If you’re an engineer, it’s not about being able to write code with AI. It’s about being able to design for how far the damage spreads when that code breaks in production.

If you’re a consultant, it’s not about introducing AI tools. It’s about figuring out which part of the work, once changed, will actually move revenue, cost, risk and customer experience.

If you run a company, it’s not about announcing an AI rollout. It’s about deciding which work to reinvest the time AI frees up into.

What survives in the AI era is not the AI operator.

It’s the designer of the whole job, AI included.

And that design always leaves human responsibility in place.

Which data to use.

Which customers to put it in front of.

Which failures to tolerate.

Who makes the final call.

What AI is good at is processing things that have already become words or data.

The unease no one has put into words yet, what customers really think, the politics of an organization, where responsibility lies, where the business should place its bets: none of that is easily automated.

Closing thoughts

The people warning about the dangers of AI are not simply exaggerating.

If anything, I think they are standing, quite early on, in a place where the ground gives way easily.

The closer your work is to AI, the more directly you’re hit by its progress. The tasks that got faster with AI are the ones that next become tasks AI can do on its own.

So the advice “learn to use AI” is no longer quite enough.

The real question comes after that.

Does the value of your work lie in using AI?

Or does it lie in the judgment, responsibility, relationships and results that remain after you’ve used AI?

If AI does take jobs, it may start not with someone far away, but with the people near AI who talk about it, sell it and use it.

🦉 I think the closer you are to AI, the more it pays to move to a slightly higher branch. Rather than running fast on sand, stand somewhere you can see which ground is starting to give way.

---

References

[1] OpenAI: Introducing ChatGPT

[2] Stanford HAI: Economy | The 2026 AI Index Report

[3] Challenger, Gray & Christmas: Challenger Report: June Layoffs Cool to 45,849, Down 53% From May; AI Leads Reasons For Fourth Consecutive Month

[4] Goldman Sachs: The Jobs AI Is Likely to Boost—and Those It May Disrupt

[5] OECD: Venture capital investments in artificial intelligence through 2025

[6] Stanford HAI: The 2026 AI Index Report

[7] Menlo Ventures: 2025: The State of Generative AI in the Enterprise

[8] Impress Watch: Japan’s generative AI use surges but still lags far behind the US and China, says Information and Communications White Paper (in Japanese)

[9] McKinsey & Company: The state of AI in 2025: Agents, innovation, and transformation