The AI Jobs Apocalypse That Wasn't

Well, well, well.

After two years of apocalyptic predictions from the very people who built the machine they said would end work as we know it, here we are.

Sam Altman, CEO of OpenAI, now says he was "pretty wrong" about AI gutting entry-level white-collar jobs. Dario Amodei, who once told the world with remarkable confidence that AI could eliminate 50% of white-collar employment, is walking it back too — reframing automation not as a destroyer of jobs but as a multiplier of output.

Both, you'll note, are eyeing approximately $1 trillion IPOs.

I'm not saying that's why. I'm just noting it.

I've been called Pollyanna.

I've been called naive.

I've been called out of touch.

Call me right.

The Experiment That Said It All

Here's the detail from the Fortune story that most people will scroll past, but shouldn't.

Altman didn't just change his mind based on data. He ran a personal experiment. He tried delegating his Slack and email responses to AI. Then he went back to doing it himself.

"We really do care about our interactions with people," he said afterward.

That's not a failure of the technology. That's a revelation about what the work actually is.

And it's the thing I've been arguing since the apocalypse crowd started making headlines: when you strip a job down to its atomic level, you don't find tasks. You find judgment. You find relationships. You find someone deciding which task matters most right now, and why, and for whom.

AI didn't want to answer Altman's Slack messages any less effectively in month three than it did in month one. He walked away because something felt off. Because the responses, however competent, weren't his. Because the people on the other end of those messages deserved more than a well-prompted output.

That's the job.

Tasks Are Not Jobs

Here's the distinction that got lost in two years of breathless coverage.

AI agents are genuinely excellent at tasks. They can draft the memo, pull the research, build the media list, summarize the meeting, prep the brief, and write the first pass at the press release. For communications professionals especially, the productivity unlocks are real.

But a job isn't a list of tasks. A job is the orchestration of those tasks.

It's knowing which task is urgent and which one can wait. It's reading a situation and deciding that the technically correct response isn't the right one. It's understanding that a particular client is going through something, and this email — the one the AI would have sent at 9:02 AM — needs to wait until Thursday when you can actually talk to them.

That layer — the judgment layer — is not a feature that gets added in the next model update. It's contextual, relational, and deeply human. You can't prompt your way to it.

Amodei's revised framing is actually more honest about this: "If you automate 90% of the job, then everyone does the 10% of the job. And the 10% kind of expands to be 100% of what people do."

Exactly. The 10% that remains is almost always the hardest, highest-value, most irreducibly human part. The judgment. The relationship. The read.

What Happens Away From the Terminal

There's another dimension that never made it into the jobs apocalypse conversation, and it matters enormously for communications professionals.

A significant portion of what makes someone genuinely good at this job happens away from a screen.

The hallway conversation that changes the direction of a campaign. The dinner that turns into a referral three months later. The relationship with a reporter that gets you a call back when everyone else gets silence. The trust that's been built over years that gets a source to go on background when they otherwise wouldn't.

None of that lives in a prompt. None of it can be delegated to an agent.

Relationships aren't just a warm feeling. They're business infrastructure. They're the reason a pitch lands or doesn't. They're the reason a crisis gets contained or spirals. And they are, almost by definition, the part of the job that AI cannot replicate — because they require presence, consistency, history, and genuine care.

The comms professionals I worry about aren't the ones using AI. It's the ones who think their value is in the deliverables. Because deliverables can be automated. Judgment and relationships can't.

The Risk Worth Taking Seriously

I want to be clear about something, because I'm not here to tell you everything is fine and you should stop thinking about this.

AI absolutely can take enough tasks from a role that an organization starts asking whether that role is still necessary. That risk is real. We're already seeing it in hiring decisions. We're seeing it in headcount conversations at agencies. We're seeing it in the way some clients are renegotiating scopes.

But that is a staffing efficiency question. It is not "AI is coming for your job."

These are very different conversations that require very different responses.

The first one says: demonstrate the value that can't be automated. Lean into judgment, relationships, strategy. Become indispensable at the layer AI can't reach.

The second one says: panic. And panic has never produced a good communications strategy.

The apocalypse crowd, whatever their motivations, did the industry a disservice by conflating the two. They made people afraid of the wrong thing. And fear, in my experience, is a terrible teacher.

Where We Actually Are

The data backs the more measured view. The Yale Budget Lab has found no significant changes in occupational mix or unemployment duration in high-AI-exposure jobs since ChatGPT launched in late 2022. Goldman Sachs CEO David Solomon — who never bought the apocalypse narrative — points to 145% civilian employment growth in the U.S. since 1962 through every major technological disruption we've seen.

History, it turns out, is not on the side of the catastrophists.

What history does show is that the people who thrive through technological change are the ones who understand what the technology actually does — and, just as importantly, what it doesn't.

AI does tasks remarkably well.

It doesn't orchestrate them. It doesn't build the relationships that give them meaning. It doesn't carry the consequence when something goes wrong.

That's still you.

It was always going to be you.