Acemoglu's 5% Example and How We Use AI at Work
Nobel economist Daron Acemoglu compares automating 5% of jobs with making everyone 5% more productive. It fits what I see: most people use AI as a search engine.
In my last post I wrote that many people around me use AI as an advanced Google search. It helps, but most of what AI could do for them stays unused.
Daron Acemoglu, the MIT economist who shared the 2024 Nobel Prize in economics, makes a related point in a long YouTube interview about AI and work. He is not against automation. His concern is companies that use AI only to replace people.
Quick Answer: Acemoglu's example: a company with 100 employees can automate 5 of the jobs at half the cost, or make all 100 people 5% more productive. In his view the second option is worth more, and the rough numbers agree: about 2.5% saved on payroll against about 5% more output. He argues that today's AI models are already good enough for tools that make workers better, but very few such tools are being built.
01.The Example
The numbers are simple. Halving the cost of 5 jobs out of 100 saves about 2.5% of the wage bill. Making 100 people 5% more productive gives about 5% more output from the same team. Acemoglu adds a second argument: experienced staff know the clients, the products and the exceptions, and a company loses that when it lets them go.
He is clear that he wants automation too. Nobody, he says, wants to go back to carrying heavy loads on their backs. The problem is a company that does nothing else with AI.
02.Why the Useful Tools Are Missing
According to Acemoglu, the AI models are not what is holding things back. Even last year's models, he says, would be good enough for useful tools for nurses, teachers, electricians or quality control. Very few of these tools exist.
He describes the reaction when he raises what he calls "pro-worker AI". Chief executives tell him it is exactly what they need. People at technology companies either nod or tell him that they are building artificial general intelligence and he lives in a different world.
He also explains why coding has moved faster than any other job. It is office work, the people supervising it know the technology, and code either runs or it doesn't. AI learns quickly when there is a clear right answer. Customer service has no such answer: every customer asks differently and there are many exceptions, which is why the AI hotlines he calls are, in his words, a nightmare.
03.What I Recognise From My Own Projects
Since November 2025 I have started six projects with AI coding tools. The consumer apps moved fastest. The ones meant for professional use are more complex and are taking longer.
Acemoglu's explanation fits that. The knowledge needed for a good tool often isn't written down. It sits with the people doing the job, and he says builders have to "go there, get your hands dirty and talk to workers". He also thinks companies should pay staff when their knowledge goes into the data an AI tool is built on.
04.How He Uses AI Himself
Acemoglu says he is "behind the frontier". He uses AI for background research and, increasingly, to criticise his own drafts and ideas. He doesn't use it for maths, which he still does by hand, and he tells his students not to let AI write for them. His worry is that people, students especially, hand over so much thinking that they stop learning.
In my view, the gap between people who really work with AI and people who use it as a smarter search engine is getting wider. Acemoglu's example puts a number on what that gap is worth to a company.
Cover image generated with Krea AI.
Sources
- Interview with Daron Acemoglu, "Could AI leave people without meaningful work?", YouTube, 2026. youtu.be/0lsasYuzhL4
- Daron Acemoglu, "The Simple Macroeconomics of AI", 2024.