Sam Altman has admitted that he overestimated the extent to which AI would displace human workers. Jensen Huang states that the argument that AI will cause widespread unemployment is completely upside down.
I would be ashamed if OpenAI was not the first large company to be run by an AI CEO.
In October 2025, Sam Altman made such a remark.
Nine months later, he changed his tune.
On July 28, on the podcast Invest Like the Best, Altman said that people "don't really want an AI CEO", because they want to know exactly who is making decisions at a company and who to hold accountable when something goes wrong.
On July 28, Altman (right) appeared on Episode 484 of the Invest Like the Best podcast, stating that people "don't really want an AI CEO".
Moreover, he admitted in May this year that he had overestimated the speed at which AI would eliminate entry-level white-collar jobs, and that the so-called "employment apocalypse" would most likely not come.
Almost at the same time, Jensen Huang dropped a line at YC's startup class: "The narrative that AI is destroying jobs is completely backwards."
Jensen Huang stated at YC Startup School 2026 that "the narrative that AI is destroying jobs is completely backwards".
One is in charge of the world's most cutting-edge AI models, the other holds core computing power resources.
Why did they both hit the brakes on the "unemployment apocalypse" theory at the same time?
Why did Altman change his mind?
Altman said people want to know who is accountable for a company's decisions, and who to hold responsible when something goes wrong.
That's why people "don't really want an AI CEO".
He added that most people to this day still trust real human beings more and prefer to cooperate with them. In other words, even if AI can make decisions, people may not be willing to hand over the company to it.
What prevents an AI CEO from taking office is not its insufficient intelligence, but the fact that it cannot be truly held accountable.
AI can calculate the optimal solution in a second, but it cannot bear the consequences of a decision in court, in the boardroom, or in front of the public.
A small personal experience of Altman's perfectly confirms this point.
He once tried to let AI reply to his Slack messages and emails on his behalf, signed "This is Sam's AI", but soon went back to replying by himself.
The reason is simple: he found that people really care about interacting with real people.
In fact, as early as the 2025 conversation where he envisioned the AI CEO, Altman had divided the roles: real humans handle external affairs, and AI handles decision-making.
What's different this time is that he drew a clearer line between "AI can make decisions" and "humans must take accountability":
AI can make decisions, but it cannot be held responsible.
Jensen Huang: The narrative of "AI destroying jobs"
is completely misunderstood by people
Jensen Huang said that people got the "AI destroying jobs" narrative backwards because they have long treated "tasks" and "jobs" as the same thing.
He put forward such a decomposition method.
A job has a clear purpose, and to achieve this purpose, many tasks need to be completed; AI can take over some of these tasks, but taking over tasks does not mean eliminating the entire job.
A job also includes communication, judgment, coordination, review, taking consequences and other elements, which will not disappear along with the part of tasks replaced by AI.
He also cited examples of radiologists and software engineers.
AI is taking over more and more of the tasks of reading medical images and writing code, but the number of people in these two positions is actually increasing.
The figures Huang provided show that software engineering positions have increased by about 10% year-on-year, and radiology positions have increased by about 20% in the past few years.
Why the growth? Because there is a huge backlog of unmet demand.
Patients queuing up in hospitals, software waiting to be developed, and piles of unprocessed demands that were impossible to fulfill before.
Once tasks are accelerated by AI, enterprises and hospitals can take on more orders and expand their scale, so they need to hire more people to do the remaining work that AI cannot handle.
Altman also mentioned a supporting fact: a year ago everyone said software engineers would be out of work, but that did not happen.
What has really changed is the job itself: you hardly write code line by line anymore, but it is still the software engineering job that everyone can recognize at a glance.
The tasks are replaced, but the job remains. But not all jobs are preserved.
Huang also mentioned that all jobs will change, and some occupations will indeed disappear completely.
The logic he stated holds true only with a hidden premise: enterprises will use the efficiency saved by AI to expand their business and hire more people.
If the market demand stays at the same level, or if the boss directly converts the dividend from AI into layoffs to reduce costs, the number of positions will still decrease.
AI has not wiped out entry-level positions
but the entry threshold is being raised
The University of Maryland and LinkUp launched a project, counting 155 million U.S. recruitment postings since 2018.
The results show that as of the fourth quarter of 2025, there is no evidence that AI has crushed recruitment demand at the overall economic level.
Top: The U.S. "AI Job Intensity" (AI jobs / total jobs) rose from 0.22% in Q1 2018 to 1.13% in Q4 2025; Bottom: "New Graduate Job Intensity" rebounded from 11.7% in Q4 2022 to 12.6% in Q4 2025, with 532,000 recruitment postings targeting new graduates in Q4 2025.
The most counterintuitive finding is the data related to new graduates.
The saying "AI will eliminate entry-level jobs first" has been widely talked about in the past two years, but this data shows that the proportion of positions explicitly targeting new graduates has not decreased but increased, from 11.7% in the fourth quarter of 2022 to 12.6% in the fourth quarter of 2025.
In the fourth quarter of 2025 alone, there were 532,000 such positions, 63% higher than in the first quarter of 2018.
Even for mathematics and computer occupations, which are supposed to be the most "massacred" by AI, the number of positions for new graduates is 55% higher than in the first quarter of 2018.
Seeing this figure, the claim that "entry-level positions are disappearing" is untenable, at least in terms of total quantity.
The report also has another counterintuitive observation:
Young, less experienced employees are more likely to gain benefits from AI tools, because AI is best at feeding massive amounts of experience to them with one click, making them "cheap and easy-to-use" early adopters in the eyes of employers.
The entry-level threshold
is getting narrower
The total number of positions has not collapsed, but that does not mean there is no cost: there is one thing that these impressive figures cannot cover up.
How did a young person grow up in the past?
They usually started with the most standardized entry-level tasks: data entry, basic coding, fundamental analysis, document sorting and so on.
These tasks are boring and repetitive, but it is by doing these tasks over and over again that people gradually accumulate experience and judgment, and develop the irreplaceable skills.
And these standardized tasks are exactly the first batch of tasks that AI takes over.
The entrance to entry-level positions is shrinking, which is the most real pressure for people who have just graduated and just entered the industry.
Your real moat
Putting Altman and Huang's remarks together, we will find that:
The more capable AI becomes, the higher the value of human work goes, shifting from "getting tasks done" to "taking accountability, building trust, and setting goals".
Machines can replace tasks one by one, but they cannot fill the gap of "who makes the final decision and who signs off on the result".
Instead of worrying "will AI take my job", it is better to ask a different question: how much of your work requires you to sign off and take responsibility? How much of it requires you to show up in person to get others' buy-in?
These parts that AI cannot replace are your real moat.
References:
https://open.spotify.com/episode/5bFBUf3X8QXaDFrve1Be5s?utm_source=chatgpt.com
https://conversationswithtyler.com/episodes/sam-altman-2/?utm_source=chatgpt.com
https://whbl.com/2026/05/26/openais-altman-says-ai-unlikely-to-lead-to-jobs-apocalypse/?utm_source=chatgpt.com https://youtu.be/I4B37S1dyQQ
Editor: Yu Yuan
This article is from the WeChat official account "AI Era", author: ASI Revelation, published with authorization from 36Kr.