David Friedberg's picture of the AI economy begins with a painter. Instead of one man painting one wall, imagine him in charge of five robots painting it. Far more homes get painted in a day, so, Friedberg said, he gets paid more for the work.
Friedberg, an investor who also appears on the All-In podcast, was a guest on The Diary of a CEO, where the host spent much of the conversation testing that optimism against what he says he is seeing in his own hiring. Friedberg's starting claim was historical: through the computing waves of the 1960s and the 1980s, when people expected a desktop machine to make fifty colleagues disappear, and back to the Industrial Revolution, "We've never seen jobs decline." Jobs went up, demand went up, and incomes rose, because the new tool let one person do more per hour.
The objection he anticipated is what he calls the fixed pie: there are only so many walls to paint. "But what always happens is that the pie grows," he said — the economy grows, people invent new products, spare capital goes into new homes and new buildings. Asked who benefits from the bigger pie, he said value had reached the entire workforce, "all the way down to the lowest level." Then the qualification: "It doesn't happen overnight." The painful stretch, in his telling, runs from the day the mainframe appears in the basement and you assume you are finished, to the day your employer's revenue is up and everyone is getting raises, bonuses and new product lines.
Revenue and costs
The host's first challenge was about pace: this wave targets cognitive work and rides on the internet. "I'm telling you, I just don't see it," Friedberg replied. At his own workplace everyone uses AI, everyone does more, and then they ask for more employees.
His explanation splits a business in two. Revenue is what people pay you for what you make; cost is what you spend to operate. Automation arguments, he said, only ever run down the cost side — you lose the customer service reps — and never account for the revenue side. His example was a materials company, whose progress is limited by the slowness of making and testing experiments by hand; AI can screen millions of candidate materials in a computer, print them, test them and automate the throughput, so the company can sell more.
The host said he could relate, and then named the limit of the comparison: both of them are founders with capital. AI has made him more ambitious about hiring in the near term. "I would caveat it by saying that there's a certain type of role that I'm now hiring much less," he said — especially entry-level roles, where he now pauses because agents and tooling can do work he would once have hired for. And what about the 21-year-old graduate, the DoorDash driver, the person assembling a part on a factory floor, for whom the leap to this new world is so large that there is a "messy middle"?
Friedberg's answer was that as new products bring new revenue, "you don't cut costs, you grow costs, and you hire more people, and you up-level, and you train people. That is what I'm seeing on the ground." He added that the United States had, he thought, just published one of its lowest unemployment rates.
The company that starts with a K
Both men had gone looking for evidence at the same place. Friedberg said he had phoned the founder of the buy-now-pay-later company he described as starting with a K, after reports that it had cut its customer service staff; the founder told him he had been misinterpreted in the press.
The host said he had put the same question to him live on air, and had the direct messages. His account of the correction: the founder had around 7,000 employees and said he would be at 3,000 by the end of last summer, and the media had turned that into a story of a company that cut, failed and reversed. "But he said, we are cutting because of AI."
Then the host played tape from an interview he dated to five months earlier. On it, Klarna's founder described releasing AI early for support, with more chats handled by the assistant and answered faster; that share had since expanded slightly. The part he said had been missed was the other half of the thesis: in a world where AI is cheap and available, "the value of human interaction will be regarded as higher. The future of customer service VIP is a human." On staffing: "We used to be about 6,000 people, and now we are less than 3,000, which is two, three years since we stopped recruiting. At the same point of time, our revenue has doubled."
That is a later account than the company's February 2024 announcement, which reported that its OpenAI-powered assistant had handled 2.3 million conversations in its first month — two-thirds of customer-service chats — across 23 markets and more than 35 languages, with resolution times under two minutes against eleven previously, 25% fewer repeat inquiries, and customers still able to ask for a live agent. The company put that volume at the equivalent of 700 full-time agents' work and forecast a $40 million improvement in 2024 profit.
Friedberg did not accept the case as typical. "This is a startup, just to be clear," he said: a startup that grows very fast and then cuts heads is a common phenomenon, and he did not know how much of the restructuring was really the AI. He also asked what those agents had been doing before — four hours a day on the phone, two, eight? "Were they truly fully employed?" The host did not know, and turned the question around: what are they doing now, and is that defensible against the same spread of intelligence? They could be lawyers. They could be doing DoorDash.
His general model stayed the same. A manager who can invest $10 and make one product a year, or $20 and make five, invests the $20. "As soon as you have the ability to get leverage on how much more productive people are, you'll put more capital towards hiring more people and scale up." He left room to be wrong: "And look, I could be wrong and all jobs go away in five years and we have 20% unemployment. I don't think that that's the case." He expects growth in media and digital content, in entrepreneurship, and in what he called IRL stuff.
What he would watch, and what the data show
Pressed on when to act, Friedberg named two numbers together: wages and unemployment. If wages are declining while unemployment climbs, "That's a red flag. I mean, that's an alert moment. And that's when we have to address this from policy. But it's very speculative right now." Switch off the advancement instead, he argued, and it happens in China, in Europe, in South America, and the United States is left behind. Tracking the consequence people fear is, in his view, the most important thing to be doing now.
The host's worry was that the signal arrives too late. He cited a Stanford study showing a 13% relative decline in employment for workers aged 22 to 25 in roles most exposed to generative AI, and asked whether a decline this fast would already be beyond the conversation by the time it showed up.
That paper, by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, has been revised; the November 2025 version reports 16% rather than 13%. The authors use ADP payroll records covering roughly 3.5 to 5 million workers a month at firms continuously present in the data from January 2021 to September 2025, excluding part-time workers, workers without job titles and those aged 70 or over, and they rate occupations for AI exposure by combining task-based estimates with classifications of how Claude is actually used. The 16% is a relative decline for 22-to-25-year-olds in highly exposed occupations once comparisons are made within the same firm and month — not a national unemployment rate. In raw terms, employment in the two most exposed quintiles fell about 6% from late 2022 to September 2025, while older workers' employment grew. The declines cluster in occupations where AI automates the work rather than augments it; salaries showed little matching divergence. The pattern held when technology firms and remote-capable occupations were excluded. The authors call this consistent with AI affecting entry-level employment, while acknowledging other influences and the lack of good firm-level data on who has actually adopted what.
Friedberg said the timing the host described was the right moment to be looking, and that he understood the mechanism in his own industry. In software engineering, young people are not getting hired. You can use AI to write a lot of code, but to be really good at that you need experience writing code, so there is a fight over senior engineers and disinterest in juniors: nobody is putting a junior in a seat to write their first line any more. "With AI, you can turn one engineer into 100. But the junior guy is not going to do that. The senior person is."
The host supplied the same story from outside software. He once would have had a personal assistant reading his email, sorting it and moving it onto a project board; now an agent crosses all his inboxes every day and pulls everything into one system. "Now it's a software job... I don't need the PA. I can just do it myself." He would have kept hiring into that team and no longer needs to. Friedberg's two questions back were what else that assistant could now be used for, and where the unspent money goes: not into a bank account, but into assets, and a productive asset employs someone. Even buying the S&P 500, he said, sends the capital to people who will employ more people. "This is why economies grow when technology shows up."
The assumption he would bet against himself on
Asked which principle in his own hypothesis could turn out to be a miscalculation and produce mass unemployment, Friedberg answered without hedging: "The choice that people will make to go to higher paying jobs, to different jobs, that they see an opportunity and they make a choice to change. Because I think that that's the key to this whole thing."
His example was the New York taxi medallion — the restricted right to drive a cab, whose price climbed toward $500,000, with Citigroup lending drivers the money to buy one. Then Uber arrived, the total number of car trips in the city grew considerably, and drivers moved: more money, their own hours, no medallion rent. "But everyone made the choice. They moved. They found, the market found its way."
The host was unconvinced that this was the right scale of leap. Going from being a taxi driver to being a taxi driver for a different firm using an app, he began, is quite a — and Friedberg cut the argument off at its root. "So I could sit here and pontificate all day with you on this. And we'd both be wrong, I'm sure. I don't know what the leap is going to be for an individual. I think we have to start to pull these stories out of the economy, what's going on."
The host said he also believes the destination is better, absent existential problems, but that the journey remains a question mark for him, especially retraining at speed. Friedberg agreed. "I'm not trying to defend it and say it's going to be a clean, smooth line," he said. People could argue all day about displacement and dysfunction along the way, or a lot of people might be smiling as they jump to new jobs. "And if people are out of work and can't find work and they're challenged in paying their bills, we have a frigging problem."
Asked what the solution then is — universal basic income? — he expected more government money in some form: "whether it takes the form of UBI, government jobs grow or some other sort of financial support." Government jobs, in his description, are not productive: paid $100, a worker there produces perhaps $60 or $80 of value rather than $120. He said that is the direction of travel, and that he does not want it. "I don't want to lose my agency. I don't want to lose my capacity, my liberty, my freedom." He feels sad, he said, for people who embrace it, because he thinks they do not realize what they are giving up — to which the host's reply was that it is being offered where nothing else is.