Doomer and booster narratives both serve trillion dollar AI valuations. The decisions actually shaping AI at work are being made in performance review algorithms and warehouse tracking software.
The AI industry is selling the public two apocalypses, and they are mirror images of each other. In one, AI eats every job on earth and triggers a permanent underclass. In the other, AI lifts humanity into post-scarcity abundance. Both poles are useful to the same people, and that utility is the construct that Guardian columnist Arwa Mahdawi argues is worth examining on its own terms. She calls it "AI absolutism," and the construct is more revealing than either of its poles.
Why does this matter now? The gap between the absolutism debate and what AI is actually doing to workers has never been wider, and the gap is where real decisions are being made. AI investment accounted for nearly 60% of US GDP growth in the fourth quarter of 2025, according to Barron's, the kind of number that justifies a sales pitch. More than 500,000 tech industry workers have lost their jobs since ChatGPT launched in late 2022, per the Guardian's own AI absolutism column, the kind of number that sells the doom pole. Both numbers are real. The story they are embedded in is the product.
Start with the booster pole, since it has the larger microphone. Nvidia CEO Jensen Huang said it plainly in a 2025 CNBC interview: "You're not going to lose your job to an AI, but you're going to lose your job to someone who uses AI." Anthropic's Dario Amodei went further in his essay "The Adolescence of Technology" published in January 2026, framing AI as a "general labor substitute for humans." These are not fringe claims. They come from the CEOs of the two companies whose market caps anchor the AI investment thesis.
The doomer pole runs on the same fuel. A Wired piece on the AI bubble catalogs the "what calamity will befall us" framing. A New York Times opinion column talks about a "permanent underclass" of workers left behind. The doomer version feels like a warning, but it has the same effect as the booster version. It treats AI's economic and social consequences as a fait accompli. Inevitable doom and inevitable utopia are the same product, pitched to different buyers.
Then the CEOs who sold the strong version started hedging. OpenAI's Sam Altman walked back his job-replacement claims in Time in May 2026: "I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened." Andreessen Horowitz's Marc Andreessen was blunter in an April 2026 SF Standard interview: overstaffed companies are using AI as a "silver-bullet excuse" to lay people off. The walkbacks landed without much media oxygen, which is itself a data point.
The economic scholars cited in the Guardian column do not line up with either pole. Columbia economist Suresh Naidu points out that software is only 4 to 6% of GDP, so even a complete AI takeover of software would not be economy-wide. He calls the technology "transformative" but says the current hype is historically excessive. UC Berkeley Haas's Martin Beraja is more direct. Studies linking ChatGPT's release to entry-level software job declines are "problematic," he says, because much of the layoff pattern reflects post-pandemic overhiring unwinding, not AI substitution. Anil Dash, the former Glitch CEO, says coding is the clearest near-term applicability domain and that broader job-replacement claims are "noise" because most outputs are subjective and hard to verify.
So if the job-apocalypse timeline is shakier than either side admits, what is AI actually being used for in workplaces right now? The Guardian column lands on this question, and it is the question worth staying on. The strongest concrete counter-example in Mahdawi's piece is not a forecast. It is a May 2026 Guardian report on AI being used to surveil and micromanage workers. Amazon, Meta, and Block have all been named in employee-side reporting that the AI productivity gains their bosses cite are "overblown," in the column's phrasing, and that the systems are being used for performance flagging, scheduling, and break-tracking.
This is the part the absolutism debate obscures. Surveillance and micromanagement are not prophecies about a future AI will bring. They are present-tense business decisions, made by managers, vendors, and procurement teams, often without worker input or legislative guardrails. They are decisions that can be negotiated, regulated, or refused. They are not the terrain of inevitability. They are the terrain of ordinary politics.
The constructive payoff is a reframe. Stop asking whether AI is good or bad for the world. Start asking who is making the AI deployment decisions, in which workplaces, under what oversight, with what recourse for the workers on the receiving end. The doom and utopia poles are useful precisely because they make the question feel unanswerable. The surveillance and micromanagement case makes the question very answerable, because the deployment is happening now, in named companies, with named tools.
A trillion-dollar narrative requires a story that the technology can "eat all the work on the planet," in Naidu's words. The actual deployment underway does not need that story. It needs compliance teams, labor contracts, and regulators who ask what the software is doing and who decided. The future of AI at work is being negotiated right now, in those rooms, and the people in them are not on the conference stages.