Every week, new claims are made that AI will transform the workplace. CEO declares revolution. One think piece predicts that millions of jobs will disappear overnight. The noise is relentless.
But when you strip away the hype, a simpler question arises. What has AI actually changed about jobs in developed countries so far? The answer turns out to be more interesting and more uneven than either side suggests.
what is true
Let’s start with what the evidence supports. AI offers real productivity gains in certain types of knowledge-based and service tasks. An experimental study found that professionals who use ChatGPT to create tasks completed tasks 40% faster and with 18% higher quality (as rated by their peers in a blind test).
Another study of over 5,000 customer service agents found a 15% increase in issues resolved per hour. An industry experiment in which we worked with management consultants to perform realistic and complex tasks found that work was completed 25% faster and results appeared to be 40% higher quality (also judged by experts in blind tests). A randomized trial involving approximately 5,000 software developers demonstrated a 26% increase in completed tasks.
AI has long been discussed as a threat to jobs and livelihoods. But what is the reality? This new series explores the impact AI is already having on different professions and how people actually feel about AI assistants.
These are not small numbers. And adoption is progressing rapidly. A US study found that nearly four in 10 workers will be using generative AI in the workplace by mid-2025. This pace of adoption has outpaced the early days of both personal computers and the Internet. Across Organization for Economic Co-operation and Development (OECD) countries, companies report an accelerating integration of AI into business functions.
So the productivity story is real, especially for text-heavy, codable tasks like legal, finance, marketing, and customer service. It’s not that much of a hype.
what is being exaggerated
However, the apocalyptic prophecies have not yet come true. Employment across OECD countries remains historically strong. A review of research-based evidence produced in the United States in early 2026 found that the rapid adoption of AI has so far caused few widespread job losses or pay cuts. Additionally, a (yet unpublished) study tracking the use of AI chatbots in Danish workplaces found essentially zero impact on revenue or time logged, even among heavy users and early adopters.
why? That’s because many jobs still require tacit knowledge, physical presence, sound judgment, and the kind of situational awareness that AI can’t yet replicate. And the hiring picture is much more uneven than the headline numbers suggest. Although the use of AI by U.S. companies skyrocketed from 2023 to 2025, the report found that fewer companies are actually incorporating AI into their operations. For example, the information sector has adopted this at about 10 times the rate of hospitality.
One economic modeling exercise estimates that AI could increase U.S. GDP by 1% to 1.6% over the next 10 years. While this is important, it is far from a transformative claim.
There remains a significant gap between productivity gains from controlled studies and actual change within organizations. In most workplaces, the revolution has not yet arrived.
What is underreported?
This is where the story becomes more important, but there are some parts that are lacking in explanation. The distributional impact of AI in developed economies is even more noteworthy. Not everyone experiences this change in the same way.
The evidence on who benefits is surprisingly consistent. Inexperienced employees can benefit the most from AI tools. One study found that AI narrowed the gap between the most and least productive staff, with the lowest-performing employees seeing the greatest improvement.
In customer service, novice agents benefited the most. Even the most experienced staff showed little improvement, and in some cases, quality declined slightly. The aforementioned industry experiment found that below-average performers improved by 43%, while top performers improved by 17%. Therefore, the greatest benefits go to the least experienced employees, narrowing the gap between top and bottom performers within the company.
That sounds like good news. But there’s a catch.
While AI may compress skills within companies, the broader labor market tells a different story. Entry-level roles are shrinking in professions exposed to AI. For the first time, routine tasks that once justified hiring junior workers, jobs that provided learning opportunities for those at the bottom, will be automated.
Economic theory has long warned that automation does not automatically or guarantee the removal of workers from tasks and the creation of new tasks to replace them. An estimated 60% of jobs in developed countries could be exposed to AI.
Everett Collection/Shutterstock
In most realistic scenarios, inequalities worsen without intentional intervention. Part of the reason is that higher-income workers hold more capital assets and stand to benefit from increased returns on AI-related investments.
Here’s a pattern that’s emerging: The AI helps those already in the door and silently narrows the door for those trying to enter.
pay attention to the right questions
Sector is important. Company size is important. Occupation is important. The transition to AI is not a single story. They unfold at different speeds and have different outcomes depending on where you are in the economy.
The debate is caught between suffocating optimism and existential dread. Both are useless. The evidence points to a more unpleasant location. It is a transformation that is real but partial, fast in some corners and stalled in others, and that distributes its costs and benefits in a way that is shaped by existing inequalities.
If productivity gains are real, the question is, who will capture them? If entry-level works are disappearing, what’s to replace them? And if the gap between companies that adopt and those that don’t is widening, then we should focus on what we’re building in response. It’s not enough just to talk about it.
