Your job isn't disappearing. It's splitting in two.

I have spent my career navigating the IT industry, but I approach these shifts through the lens of an economist. When CEOs start warning that AI will “wipe out half of all entry-level jobs,” I don’t just hear a sales pitch; I see a structural claim that needs to be interrogated against the data. This post is my attempt to reconcile the hype with the economic reality.

The CEOs warning loudest about AI taking jobs are not primarily speaking to workers. Dario Amodei at Anthropic predicts AI could “wipe out half of all entry-level white-collar jobs” within five years. Sam Altman says the impact “will begin to be palpable.” Demis Hassabis told Davos the AI transition will deliver ten times the impact of the Industrial Revolution at ten times the speed. Three competitors, one narrative. As a technology sales specialist, my first instinct is to ask who else is in the room — and in this case, the answer is clear: capital markets. Ben Goertzel, who coined the term AGI, puts it bluntly: if all jobs are going to be taken over by AI, you better own a piece of it. The job-displacement warning and the investment pitch are the same sentence.

That framing does not mean the underlying risk is invented. It means the claims are doing two jobs simultaneously, and conflating them produces bad analysis. Strip away the investor-signaling function, and what is the actual evidence for AI-driven displacement? Less settled than the headlines suggest, more sector-specific than the general alarm implies, and almost certainly playing out along a fault line that nobody in a boardroom is eager to draw attention to.

What the data actually shows

Pew Research found that 63% of American workers say they rarely or never use AI in their jobs — and this was measured after two full years of mainstream ChatGPT availability. That figure sits uncomfortably alongside the vendor narrative. If AI is everywhere, it is not showing up in self-reported worker experience at anything close to the penetration rate the press releases imply.

What is showing up is real but narrow. Salesforce eliminated roughly 4,000 customer support roles, a decision Marc Benioff attributed directly to AI. That is a named company, a named CEO, and a specific job category — verifiable and enterprise-scale. It is also, notably, not a knowledge-worker displacement story. Customer support at Salesforce’s scale involves high-volume, low-variability interactions: exactly the conditions under which today’s AI performs reliably. The parallel case is not software engineers; it is the offshore call-center wave of the early 2000s. Which raises the question that gets less attention than it should.

The wrong jobs are in the headline

Alex Imas and Soumitra Shukla, economists at Chicago Booth and Harvard Business School respectively, argue that the dominant AI exposure indices — the ones that have generated dozens of alarming data visualizations — are measuring the wrong thing. The Eloundou et al. 2023 paper that launched the “80% of US workers are AI-exposed” finding defined exposure as potential time-reduction, not displacement probability. Those are different claims. A radiologist whose diagnostic notes can be drafted by AI faster than she writes them is “exposed” in the Eloundou sense while facing zero displacement risk — if demand for radiology reads is growing and her diagnostic judgment remains the value-creating step.

What actually determines whether a job disappears, Imas and Shukla argue, is job dimensionality: the number of distinct, complementary tasks that together constitute the role. A truck driver’s job is dominated by one core function. A management consultant’s job has seven or eight distinct complementary tasks, and automating any one of them tends to make the remaining ones more valuable, not less. This explains why 870 FDA-approved AI tools are now in radiology and 66% of doctors report using at least one, while the number of radiologists has not declined. The automation absorbed the low-value documentation work; the judgment work got more prominent.

The occupation that should be generating more alarm than software engineering is trucking. Aurora Innovation and Kodiak Robotics are already running commercial autonomous pilots on constrained long-haul routes. Three million American workers, predominantly non-college, concentrated in communities where driving is the economic backbone. The displacement mechanism there is not augmentation but substitution, and the timeline is not theoretical. Yet trucking gets a fraction of the coverage of software engineering — because the people writing the coverage know software engineers, not truckers.

The jobs that split

Here is what I have actually watched happen — not in press releases, but in the work itself. My own market and competitive intelligence function, as part of a specialist sales organization, is unrecognizable from two years ago.

My role has effectively split in two:

• Commodity production. Tasks like sifting through news, tracking market movements, and synthesizing analyst commentary used to consume most of my cycle. Today, agents handle this overnight.
• Judgment-heavy tasks. This is where the value now lies. Interpreting what a signal means for our competitive position, validating vendor claims against procurement reality, and deciding which questions actually matter — this work has grown.

The output volume has doubled, yet the team has not. The job hasn’t disappeared; it has evolved. While commodity tasks have been automated, the judgment-heavy work remains — and it is significantly harder to hire for.

Who will and won’t adapt

The honest division is not between “AI-proof” and “AI-exposed” jobs. It is between workers whose employers have the organizational capacity and incentive to restructure roles around AI, and workers whose employers simply replace them with the tool. Luke Michel, a 68-year-old content strategist at Dana-Farber Cancer Institute, took an early retirement package rather than reskill. His quote is worth sitting with: the time and energy required to learn a whole new vocabulary and skill set, he said, was not worth it at his stage. That is not a failure of ambition. It is a rational calculation that the transition costs exceed the remaining career horizon. He will not be the last person to make it.

The “someone who knows how to use AI will replace you” formulation — which has become the industry’s default response to displacement anxiety — shifts responsibility to individual workers in a way that is convenient for everyone except the workers. It is true as far as it goes. It does not go far enough. Reskilling at 50 is not the same as reskilling at 25, and the structural reallocation that labor economists model across decades does not feel like a growth opportunity at the individual level.

Alex Imas’s argument regarding the “relational sector” — some 50 million US workers in care, education, hospitality, therapy, and artisanal craft — offers a genuine structural counterpoint. His experimental evidence is striking: human-made artwork gained 44% in value from exclusivity cues, while AI-generated artwork gained only 21%, as the mere involvement of AI deflated the premium. Starbucks reversed its automation push partly because customers signaled that handwritten cup notes and ceramic mugs were worth more to them than faster throughput. Whether it’s the CEO of Anthropic booking a human financial advisor instead of a robo-adviser, or the line of customers willing to pay extra for a barista experience rather than a vending-machine equivalent, people are revealing something real about where “human-made” retains its premium.

The catch is that relational-sector jobs are predominantly lower-wage, often part-time, and lack the career ladders of the knowledge-worker positions being disrupted. Moving from loan processor to barista or home health aide is structural reallocation in the economic sense. It is a pay cut and a loss of professional identity in the human sense. Labor-market models can accurately predict where jobs will go; they fail to account for the personal, financial, and emotional pain involved in that transition.

Why the “AI doom” narrative misses the point

Amodei is directionally right about white-collar displacement — but wrong about the mechanism. The scenario that materializes will not be mass unemployment so much as a bifurcation of the workforce: a stratum of workers who used AI to multiply their output and became genuinely irreplaceable in their organizations, and a stratum who either could not access the tools, could not restructure their roles fast enough, or worked for employers who decided substitution was cheaper than augmentation.

The divide will not track education cleanly or geography cleanly. It will track organizational capacity: whether your employer has the management competence and cultural openness to reorganize work around what AI can do. Large enterprises in regulated industries will be slower but more deliberate. Small employers will be faster but more brutal (though Oracle and Block have been counter-arguments). The government labor market will be last and least prepared.

Enterprise AI adoption is genuinely faster than cloud or mobile by most meaningful measures. Anthropic’s Claude Opus can already complete roughly four hours of focused human work with 50% reliability — a benchmark that will improve every quarter. But Goertzel’s organizational-friction observation is the most underrated constraint in this entire debate: even when 90% of a job function could technically be handled by AI today, organizations are slow at reshuffling how work actually flows. The technology lead-time over the deployment curve is longer than the hype assumes. That gap is where workers have time to adapt — if they use it.

The people who will look back on this transition most bitterly are not the ones whose jobs disappeared overnight. They are the ones who had two or three years of signal that their role was splitting, did not restructure it themselves, and discovered too late that their employer had done it for them.