How a New “Moore’s Law for AI agents” Could Help Predict Progress
Also: Startups growing faster, a discovery in two days that took humans ten years, and more.
It has been a bit more quiet week in AI than what we’ve been getting used to in 2025, but there was still some interesting news, and here we’ll take a closer look at a new “Moore’s Law for AI Agents”. The framework tries to predict the progress in the field by comparing 13 different models over a six-year period seeing a pattern which they hope to aid people in understanding what’s ahead.
Number of the Week
$1 trillion. The expected amount invested in data centers in 2028, according to research by Dell’Oro Group and highlighted by Jensen Huang, CEO of Nvidia, at the company’s GTC conference this week. [Source]
How a new “Moore’s Law for AI agents” predicts progress
The iconic “Moore’s Law” has contributed to setting the pace for the chip industry, and now a research center has presented an equivalent for AI agents.
METR (Model Evaluation & Threat Research) is a non-profit based in California with the mission of developing “scientific methods to assess catastrophic risks stemming from AI systems’ autonomous capabilities and enable good decision-making about their development.”
The team, who has top AI researcher Yoshua Bengio among their advisors, are now proposing a framework which compares how long a human would take to do a certain task versus how long it would take an AI to complete it with at least a 50% successful rate.
“Current models have almost 100% success rate on tasks taking humans less than 4 minutes, but succeed <10% of the time on tasks taking more than around 4 hours,” they write in a blog post presenting the framework.
For the last six years, AI’s agentic capabilities, they find, have approximately doubled every seven months.
“If these results generalize to real-world software tasks, extrapolation of this trend predicts that within 5 years, AI systems will be capable of automating many software tasks that currently take humans a month,” the team writes in a paper.
A lot of the progress in the field has happened in 2024 and 2025, they recognize, and if that rate continues, it will shorten the estimate by about 2.5 years.
Startups growing faster thanks to AI
Y Combinator, the renowned startup accelerator behind successes such as Airbnb and Dropbox, is seeing its most profitable fund in the organization’s 20-year history, attributing it to AI, CEO Garry Tan told CNBC.
Their latest batch of startups have been growing in aggregate at 10% per week.
“That’s never happened before in early-stage venture,” Tan said.
The explanation is in the efficiency gains that AI can offer, especially in coding. For about one in every fourth startup, 95% of the code is written by AI, they estimate.
It means that the number of engineers required to make something that takes off has been dramatically cut. Y Combinator is seeing companies with $10 million in revenue with teams of less than 10 people.
And AI is also on the customer-facing side of the companies — around 80% of them are what they call AI-focused with some robotics and semiconductor startups as well.
AI discovered in two days what took scientists ten years
After ten years of research, José Penadés, a professor of microbiology at Imperial College London and his team made a breakthrough in antimicrobial resistance (AMR), a phenomenon that causes more than a million lost lives a year.
Shortly after, the university was contacted by Google to test out their new “AI co-scientist” platform, and Penadés decided to test it out by giving it the same scientific questions that they had explored and where the answers were not yet published.
After just two days, the system came up with the same result as the human researchers.
“The algorithm was able to look at the available evidence, analyse the possibilities, ask questions, design experiments, and propose the very same hypothesis that we arrived at through years of painstaking scientific research, but in a fraction of the time,” Penadés said.
Mary Ryan, professor and vice provost for research and enterprise at Imperial College London, is optimistic of the potential of the technology to assist in addressing other complex challenges as well like pandemics, environmental sustainability, and food security.
“To address these urgent needs means accelerating traditional R&D processes and Artificial Intelligence will increasingly support scientific discovery and pioneering developments,” she said.
Image of the Week
Google has launched its first satellite in the FireSat initiative, which should improve wildfire detection. Using AI, FireSat will compare what it sees on the ground with previous imagery to determine if there is a fire. When the system is rolled out, it should be able to detect fires as small as 5 x 5 meters, where earlier satellites couldn’t before they had reached two to three acres in size, according to Google.
Exciting news out there?
In creating memes, AI scored higher than humans on average, measured by creativity, humor and shareability. However, the best humans still performed better than the AI [Source]
Feedback?
Thanks for reading, and feel free to respond directly with suggestions for improvement, tips, and anything in between :)