AI Is Now Out-Working the People Who Build It
DEV Community ● Covered by 5 sources
AI labs say their systems are now doing more research work than the humans running them. That head start is showing up in money, medicine, and politics.
Based on reporting by DEV Community — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Inside frontier AI labs, the quiet shift is no longer theoretical: internal numbers show coding agents are now generating about three full workdays of research output for every workday a human researcher contributes. Token use per researcher has jumped 124-fold since December, the typical researcher is spending more than $600 a day on agents, and the heaviest users are well past $7,000. About eight in ten researchers are running four or more agents at once. The lab behind those figures says it has reached the “automated research intern” milestone its leadership had set for this month, while a fully automated AI researcher is still penciled in for early 2028.
And this isn’t isolated. Another major AI company says its autonomous research system just finished 8th out of 4,000 teams in a competition focused on getting a model to reason better. That’s a flashy result, sure, but the bigger point is colder: AI is starting to improve AI at a level the company is happy to compare with human experts. That is exactly why these labs keep their best unreleased models under lock and key. The advantage compounds quickly, and now it can be measured in workdays and dollars instead of breathless demo language.
Medicine got a more tangible signal too. A drug designed by AI from scratch — target chosen by the system, molecule drawn by the system — was originally aimed at a lung-scarring disease. New trial data suggests patients treated with it looked biologically younger across six separate aging-clock models, and one analysis put the drop at nearly three years. The trial was small, with 42 patients, and the dose that showed the strongest anti-aging signal was not the dose that worked best on the lung disease itself. That makes the result interesting in two different ways, not one.
Public sentiment, though, is still stubbornly sour. A new national poll found 70% of Americans are more worried than excited about AI, and that unease cuts across party lines. More than half of adults now use AI regularly, up six points from last year, but only 18% say they trust AI-generated information most of the time. Seven in ten think AI is already taking jobs, opposition to local data centers is nearly universal, and 44% say they trust neither party to handle AI policy well. So while the industry keeps collecting technical wins, the public is busy treating the whole thing like a bill that hasn’t arrived yet.
My take — AI-written commentary, not fact-checked reporting
This is what real AI progress looks like: less demo theater, more internal leverage, and a lot more electricity bills. The industry loves talking about capability; the public is mostly staring at jobs, data centers, and the people who will get blamed when the model hallucinates an overnight hike into a rescue call. That gap is where the politics live now.
Read more about this at: DEV Community