Dear Aventine Readers,

It's time to take stock of 2025! At the start of this year, we made some predictions about what would happen in the fields of science and technology. In this week’s newsletter, we take a look at what we got right — and wrong — in a year of significant breakthroughs, disruption and change. 

We also take a look back at the advances we wrote about throughout the past 12 months and highlight 10 that we believe will have the most lasting impact on the future.

As usual, thanks for reading.  We are going to pause the newsletters over the holidays and be back in early January. 

Happy Holidays! 

Danielle Mattoon 
Executive Director, Aventine

Subscribe

Subscribe to our newsletter and be kept up to date on upcoming Aventine projects

A little over a year ago, we made some bold predictions about what would happen in science and technology during the course of 2025. Now, as the year draws to a close, we thought it would be interesting to revisit those forecasts to see how they played out. Here’s what we got right — and what we got wrong — about the last 12 months.

AI agents will take on more work

AI agents absolutely showed up for work in 2025, with tech companies launching hundreds of them. Microsoft has agents to whip Word or Excel documents into shape. Cognition’s Devin platform is an AI software engineer that undertakes entire projects. OpenAI baked agents into ChatGPT to help people automate tasks. These tools appear to help streamline tasks; less clear is whether they’re actually improving productivity. Results on that are in their infancy. One assessment by METR, a nonprofit that evaluates AI capabilities, found that AI tools slowed down software developers. And adoption inside companies has been slower than anticipated too, over concerns around reliability, accuracy, privacy and security. So agents may be taking on work, but we don’t yet have a clear idea of how useful that is.

Robots will get way smarter

Oh yes. Advances in AI, borrowed from large language models, vastly accelerated robotics progress. Google DeepMind showed that its Gemini Robotics system can independently perform complex, intricate tasks such as packing (basic) lunchboxes or making origami models. Physical Intelligence, a robotics startup, demonstrated that its robots can operate in unfamiliar and messy environments, like an apartment. And Figure, a company that makes humanoid robotics, trained one of its machines to work at almost human speed sorting packages in a warehouse environment. Yet sustained and reliable execution of a wide range of tasks still isn’t possible, meaning that these advances remain largely confined to the lab for now.

Quantum computing will have chance to prove itself

Getting there. Once considered an esoteric branch of physics, quantum computing has entered the realm of the possible — that is, if certain engineering challenges can be overcome. This would pave the way for devices that could one day revolutionize fields such as drug discovery and battery science. A key result, published by Google at the end of 2024, showed the technology’s biggest stumbling block — susceptibility to errors — was surmountable. Additionally, Microsoft developed a new quantum chip that could lead to more reliable, scalable quantum computers. And IBM laid out a roadmap for building the world's first large-scale quantum computer, due in 2029. As 2025 Nobel laureate John Martinis told Aventine earlier this year, what remains is just the engineering challenge of “closing the gap” between current hardware and a device that powers commercial applications. So it’s now a question of when, not if.

Nuclear power will continue its renaissance

The nuclear love-in intensified in 2025 and shows no signs of flagging. Formerly decommissioned plants are well on track to coming back online: The Palisades nuclear power plant received an operating license from US Nuclear Regulatory Commission, the first ever for a previously decommissioned plant, and it’s hoped that Three Mile Island may restart as soon as 2027, about a year ahead of schedule. The building of new fission plants continues, particularly in Asia, where new reactors in China and India came online this year; China alone approved 10 new plants in 2025. Then there’s fusion: Experiments confirm that the science underlying the technology is now largely proven, though admittedly with expensive and high-risk engineering problems left to solve. And the easing of US regulations to accelerate deployment of nuclear power means that the renaissance should continue to flourish.

AI will enable more scientific discoveries

Absolutely. AI is rapidly transforming all sorts of facets of scientific discovery. A new research tool developed by Google helped scientists develop hypotheses and propose research plans; in one case, it correctly proposed how existing drugs could be repurposed for the treatment of acute myeloid leukemia. A model developed by Stanford University and the nonprofit biomedical research organization Arc Institute designed new viruses capable of killing E. coli, a small step along a very long road toward AI fully synthesizing new forms of life. And models from Google DeepMind and OpenAI achieved gold-medal standard at this year’s International Mathematical Olympiad, hinting at a near future where AI can discover complex mathematical proofs. AI is making ever larger contributions to science, and looks set to make ever more significant ones in the future.

AI scribes will join your medical team

If you’ve seen a doctor this year, you’ve likely seen this time-saving trend take hold. According to an American Medical Association survey, 66 percent of surveyed physicians report using an AI scribe five or more days a week. A trial carried out by The Permanente Medical Group shows the technology is having a big impact, saving the company 1,794 working days — nearly five years of work hours — in one year, while 47 percent of patients said their doctor spent less time looking at their computer during consultations. There are still hangups around privacy and integration with electronic health records, and it’s unlikely doctors will be freed from checking the work of the AI. But the trend is clear: AI scribes are part of the medical mainstream.

Cancer vaccines will become a reality

Sadly not. Moderna’s suggestion of a fast-tracked mRNA-based cancer vaccine launching in 2025 didn’t materialize. While clinical trials have continued, the Department of Health and Human Services under Robert F. Kennedy Jr. canceled several mRNA vaccine programs in 2025, which appears to have chilled the mood for all treatments that use that technology. The first results from Moderna’s mRNA-4157 trial — which tests a customized vaccine for melanoma that teaches the immune system to launch a strong response against the cancer cells — are now expected in late 2026. Cancer vaccines are still an exciting scientific prospect, but their arrival is taking longer than we hoped.

Geothermal will get hotter

Yep, literally. One startup pursuing advanced geothermal power, Fervo Energy, drilled a hole more than 15,000 feet deep where temperatures exceed 500°F — well above regular geothermal and on track to provide greater power and efficiency. Meanwhile another startup, Quaise Energy, which develops high-energy microwave drilling systems that can vaporize rock and drill holes more than 10 miles deep, carried out its first field trials. Residential geothermal advanced, too, with Dandelion Energy, a startup, announcing a partnership with home construction company Lennar Corporation to build 1,500 new homes with geothermal heat pumps. And political will remains strong, with close links to oil and gas technology helping geothermal dodge hostility toward renewables faced by wind and solar.

Crypto will have its day in the US

It was crypto’s year — until November. Digital assets activity jumped 50 percent in the US in the first seven months of 2025 year-over-year, with over $1 trillion of transactions. A big part of the success was the approval of exchange-traded funds, or ETFs, based around the cryptocurrencies Bitcoin and Ether, which made crypto investing more accessible. In the middle of the year, the president also signed the GENIUS Act into law, paving the way for the first federal stablecoin, a cryptocurrency pegged to the US dollar. Meanwhile, the SEC shifted its crypto regulatory approach from enforcement heavy to one focused on building clear frameworks through rulemaking. But as fall turned into winter, crypto crashed, with Bitcoin losing all of its gains for the year. While cryptocurrencies are more mainstream than they’ve ever been, they remain volatile and high-risk. 

Decentralized AI could be the answer for data protection

Not quite. There has been some notable progress in this area, the biggest of which is that Vana, a decentralized AI startup, created a fully user-owned network that allows individuals to upload data and govern how it is used, voting on which models can be trained on their data and receiving a proportional share of their ownership. But trust in the concept of these systems — whose AI models are built on distributed computing networks where nodes work collectively to provide computing power — was undermined after two major security incidents hit providers of the technology. While the community that has championed the approach remains committed to the idea, progress is far slower than we expected.

(We asked ChatGPT to grade our predictions, and here’s what it said: “Overall grade: B+ Strong thematic instincts, timelines that were a touch optimistic.”)

Every month, we round up a selection of the latest advances that look set to shape our futures. Here are ten of the most impressive breakthroughs from the last twelve months.

1 - Gene therapy slowed Huntington's disease. In a study out of University College London, 17 patients received a new genetic treatment called AMT-130, which targets an error in a patient’s DNA and involves a 12- to 18-hour brain surgery. Three years later, disease progression was on average 75 percent slower based on measurements of cognition, motor skills and daily functioning compared with expected decline. UniQure, which developed the treatment, plans to apply for US regulatory approval in 2026, though it could take up to five years to obtain licensing.

2 - AI almost entirely solved translation. Meta’s latest AI translation model, SeamlessM4T, can perform real-time speech-to-speech translation from 101 languages into 36 target languages and is up to 23 percent more accurate than previous speech-to-speech systems. 

3 - Scientists brought extinct wolves back to life, sort of. The dire wolf — a large canine that went extinct about 13,000 years ago — has been recreated by a company called Colossal Biosciences. To do it, they sequenced the DNA of dire wolf fossils, identified key features in the genome relating to characteristics like size, build and coloring and then edited the DNA of a regular gray wolf to include those features. The results look like a dire wolf, but lack the full genetics of one. It could be a path toward conserving endangered species. 

4 - Big Pharma found a new drug development pipeline: China. In May, Pfizer paid $1.25 billion upfront for the rights to license an experimental cancer drug from China's 3SBio, a deal that could ultimately be worth as much as $6 billion. This is one deal of many, reflecting China’s transformation from a copycat producer of generic drugs to a major player in drug development.

5 - China's DeepSeek made advanced AI affordable. In January, a new AI model, R1, from Chinese tech firm DeepSeek offered performance on par with OpenAI and Anthropic. Most impressively, DeepSeek used clever engineering to overcome a lack of resources, dramatically reducing the computational power — and spending — required to train the model. The news was widely regarded as a “Sputnik moment” for the US, showing that innovative, low-cost overseas competitors can challenge the front-runner. 

6 - Women’s skin cells can be turned into eggs. Women who don’t produce eggs as a result of age, illness or a medical treatment cannot give birth to a genetically related child. Now, researchers at Oregon Health & Science University have shown that it’s possible to turn skin cells into fertilizable human eggs by transplanting the nucleus of a skin cell into a donor egg whose nucleus had been removed. The procedure is still highly error-prone, but hints at an answer for women who can't produce eggs but desire a genetically related child. 

7 - A new bioelectronic implant treats rheumatoid arthritis. The inch-long SetPoint System is surgically implanted in the neck and delivers a one-minute daily electrical stimulation to the vagus nerve, a key part of the nervous system that links the brain to major organs. In a yearlong randomized controlled trial involving 242 patients, the therapy helped more than half of recipients achieve remission or significantly reduced symptoms without drugs. 

8 - Google's AI Scientist made real research advances. A new Google research tool based on a collection of AI agents can generate scientific hypotheses and assess them for viability, novelty and testability, ultimately coming up with a research plan with a list of ideas for scientists to test. Early results are promising. The tool discovered that an existing molecule, KIRA6, could be repurposed for the treatment of acute myeloid leukemia. And in just days it proposed a novel approach to antimicrobial resistance that had taken humans years to develop. 

9 - An AI designed bacteria-killing viruses. In the first example of AI writing usable sequences of DNA that make up an entire genome, researchers used artificial intelligence to design viruses capable of killing E. coli. Scientists first trained the AI on the genetic sequences of around two million known viruses, then used the system to generate thousands of candidate sequences, 16 of which could infect and destroy E. coli bacteria. It is a small step along a very long road to AI fully synthesizing new forms of life.

10 - Robotic surgeons successfully removed organs from pigs. An AI-powered robotic surgeon developed at Johns Hopkins autonomously removed gallbladders from eight pig cadavers with a 100 percent success rate. To train the system, researchers fed it 17 hours of video showing 16,000 individual motions performed by human surgeons on pig cadavers. The promise is that robots trained on procedures performed by the world’s best surgeons could someday provide expert care in regions far beyond the geographic reach of today’s specialists.