Ex-Google Insider Mo Gawdat Says Your Job Has About Three Years

Mo Gawdat would like you to know that 2027 is coming. Not as a vague threat. As a deadline. The former Google X executive, who has built his post-Silicon-Valley identity around being the guy willing to say the quiet part loud about AI, sat down with Steven Bartlett and delivered the kind of forecast that makes a twenty-two-year-old with a law degree want to lie down on the floor. His argument is not subtle: entry-level knowledge work is already hollowing out, the middle is next, and the notion that blue-collar jobs are the ones in danger is, in his telling, exactly backwards.

The carpenter is fine. The call center agent is not. Gawdat draws the pyramid clearly: manual laborers at the base, mundane knowledge workers above them, then the analysts and paralegals, then leadership. Most people assume automation eats from the bottom up. Gawdat flips it. Physical dexterity is hard. Clicking around on a computer is easy. And, he notes with the mild embarrassment of someone catching himself in a confession, he is probably one of the people being paid to click around on a computer.

There’s a lot of people that are paid in the world to click around on a computer. I’m probably one of them, to be honest.

Labor Arbitrage and the Capitalism Problem Nobody Wants to Name

Gawdat is at his best in this episode not in the job-by-job breakdown, which any McKinsey deck could give you, but in the structural argument underneath it. He asks Bartlett to imagine capitalism without labor arbitrage. The whole machine, he says, runs on using labor and capital to produce things cheaper than you sell them. Remove cheap labor from the equation and you do not get utopia. You get a world where workers cannot afford to buy the things that the machines are now making. You do not need 100 percent displacement for the economy to spiral. Ten or twenty percent is enough. At that point, he says, you are not watching a disruption. You are watching something closer to a prelude.

It’s not that jobs will end first. It’s that productivity gains will make businesses not want to have as many people, costly humans, costly emotional humans when the job can be predictably done for cheaper.

He also makes a point about specialized robots that actually lands. Everyone is obsessing over humanoids. Nobody is talking about self-driving cars, which are already robots, already deployed, already winning. The Boston Dynamics dog is more efficient on a battlefield than anything built to look like a person. The humanoid form factor is a PR choice, not an engineering one. When Bartlett mentions the Figure AI video of a robot sorting packages for eight straight days, then walking over to charge itself and coming back, Gawdat does not flinch. That is the blue-collar disruption people assumed was decades away.

On Sam Altman’s Convenient Memory Loss

The best run of the episode belongs to Bartlett, actually, who walks through Sam Altman’s greatest hits on AI job destruction with the energy of a prosecutor reading back a defendant’s prior statements. In 2015, Altman said his job was to help people destroy jobs. In 2023, full stop, jobs are going away. In May of this year, he said he was wrong about the pace of white-collar disruption and was delighted to be wrong. Gawdat’s read on the reversal is not charitable, and he does not pretend it is.

The script as you rightly said had an objective and a target either way. And Sam Altman, I used to say the Altman is a brand, it’s not a name.

His point is not that Altman is uniquely villainous. It is that the incentive structure produced the message, and the message changed when the old one became a liability. People are booing at commencement speeches. They are attacking data centers. The fear that the tech industry spent years cultivating to drive adoption has metastasized into a political problem, and now the same voices that said full stop are saying actually, maybe not so fast. Gawdat’s preferred test for credibility is simpler: watch what people do with money. Anthropic declined a $500 million contract over human targeting and surveillance. OpenAI took it. That tells you something.

You have to start observing who’s actually behaving in a way that is making AI work for humanity, and who is behaving in a way that is making AI work for their share values.

Gawdat’s politics are a bit of a detour, and not a short one. He declares democracy effectively over, gestures at unspecified video evidence of crimes without arrests, and says he chooses not to talk about politics until his next book. That is a lot of politics for someone not talking about politics. But the underlying anxiety is real and not easily dismissed: governments that struggled to keep people financially afloat during COVID are not remotely prepared for structural unemployment at scale. If 2027 is when this starts to bite, that is not very much runway.