Dear Aventine Reader, 

The idea that artificial intelligence would someday be effective and scalable enough to take large swaths of mundane office work off our hands has been both a promise and a threat for decades. Now, with the advent of generative and agentic AI, it’s edging closer to reality. 

AI systems can code, write and summarize better than many recent college graduates, and certainly do it all faster. They can plan menus and organize travel. They can sift through a volume of résumés that would be impossible for a human to tackle. And although these are early days, there are signs that AI’s new abilities could already be affecting the labor market, especially in entry-level office jobs. 

Historically, new technologies displace some jobs and create others. Will that be the case with AI? Aventine spoke with AI experts, economists, and labor market researchers to get a better sense of what might be coming. They are in agreement that disruption is ahead, and that we’ll see new, currently unimaginable, jobs emerge. But how long the disruption lasts, and who gets disrupted, will depend a lot on how governments, companies and workers manage it. 

Also in this issue: 

Thanks for reading, 

Danielle Mattoon 
Executive Director, Aventine

Subscribe

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

Annabel Beales, a copywriter from Southampton, England, had her dream job at a garden center, interviewing experts and writing about growing vegetables and tending plants. By late 2023, though, the work had slowed. She explained to The Guardian that she overheard her boss muttering, “Just put it in ChatGPT.” Shortly before Christmas that year, she was let go. Today, she says, the company’s website content is largely AI-generated. 

Beales is hardly alone. As more people engage with the latest AI models, which can perform increasingly complex and non-routine tasks, they are exploring how the technology can speed up or automate work in their own domains. Professionals around the world are now wondering whether their teams could or should shrink, particularly among the junior ranks where more work tasks may be automatable. 

Fears are being stoked by proclamations from the leaders of top AI companies. Dario Amodei, the CEO of Anthropic, has said that AI could soon eliminate 50 percent of entry-level office jobs and push unemployment rates as high as 20 percent in the next five years. Sam Altman, the CEO of OpenAI, has predicted “whole classes of jobs going away,” with the 2030s likely feeling “wildly different from any time that has come before.”

While there is undeniably a promotional dimension to the comments of Amodei and Altman — they’re touting the power of technologies they profit from selling — it seems increasingly clear that AI will upend the labor market in a material way. What we don’t know is what this might look like and how long it will take. 

“Workers and companies are going to be entering a period of likely unprecedented new disruptions,” said Mark Muro, a senior fellow at the Brookings Institution who focuses on the impact of AI and innovation on the economy. “The question is, how long that lasts and how they manage them.”

Aventine spoke with a variety of AI experts, economists and labor market researchers to understand this dynamic. The consensus view is that while AI is certain to reshape the labor force in ways that will be painful for many, the most likely long-term outcome is that the labor market will readjust as new forms of work emerge over time. There is less consensus about exactly how protracted and painful that process will be — factors that will be influenced by the way companies deploy AI, how the economy responds to what the technology can do and what measures governments put in place to ease the process.

Job loss by the numbers

There is no shortage of anecdotes about AI encroaching on white-collar work. The richest vein relates to software development, with recent stories from The New York Times and Wired describing how programmers are forced to work faster, often in smaller teams, to produce the same quantities of work on far tighter deadlines now that AI can produce such high-quality code. Over the last year or so, those engineers say, their work has become more focused on reviewing code and identifying bugs in work created by AI than on writing original code.

But programming is the tip of the iceberg. In early May, IBM’s chief executive, Arvind Krishna, said the company had replaced the work of several hundred human resources workers with AI agents. JPMorgan's CEO of consumer and community banking, Marianne Lake, has said AI will allow the bank to reduce head count by 10 percent in its operations and account services departments. In a memo, Shopify's CEO told its managers that they must prove that any prospective new employee's work can't be done by AI. Amazon CEO Andy Jassy predicts his company’s corporate workforce will shrink “in the next few years” because of AI.

These trends have yet to play out in a meaningful large-scale way. While US unemployment is at 4.1 percent — up from 3.7 percent in August 2022 — that number is low by historical standards. Zoom in and things get a little less comfortable: Among recent graduates, age 22–27, whose work may be more affected by AI, the unemployment rate had risen to 5.8 percent by March 2025 from 4.2 percent in August 2022, an increase of 38 percent compared with 14 percent for all workers. Entry-level hiring in the US as of April 2025 was down 23 percent since March 2020 according to LinkedIn, a steeper drop than the 18 percent fall in overall hiring. That said, these changes could be attributed to factors beyond AI, such as a correction from overhiring during the Covid-19 pandemic, said Zanele Munyikwa, an economist at Revelio Labs, which analyzes workforce data.

Directly linking AI to these shifts is tricky, but there are hints that the technology is affecting hiring. Munyikwa issued a report showing that since November 2022 there has been an approximately 31 percent decrease in the volume of job openings for roles that are highly exposed to automation by AI compared to a 25 percent decrease in roles that are less exposed. “Compared to maybe what we might expect given the current state of AI capabilities, the amount of shift is pretty small, pretty gradual,” said Munyikwa. She predicts that before significant changes show up in national economic data, junior workers may find salaries squeezed as employers weigh human costs against automation.

Big shocks are on the radar

As the abilities of large language models have begun to reveal themselves, researchers now have a better sense of which tasks AI can automate than they did when ChatGPT first burst onto the scene in 2022, and they can map these tasks against occupations . A 2024 Brookings study estimated that more than 30 percent of US workers could see at least half their tasks affected by AI, while for 85 percent of workers at least 10 percent of tasks could be exposed. At a more granular level, the study found that some roles, such as bookkeeping and insurance underwriting, are made up of tasks that could be entirely replaced by AI. Other jobs likely to face extreme disruption include secretarial work, customer service operations and travel agents.

The difficulty in predicting how exposure to AI ripples through the workforce lies in the fact that, for the most part, AI can’t automate away an entire job. Instead, it can automate specific tasks that form part of a role — say, composing emails, writing code or analyzing data — but usually not all of them, leaving an important part of the job to a human. 

There are two main ways this can play out, with decisions likely to be influenced by sector, the size of a company and the kind of work it does. In one scenario, companies automate tasks and then use a small number of employees to perform the remaining human labor. In another, employees are augmented by AI, with the technology freeing up their time to handle new or more complex tasks. Pure automation leads to job loss. Augmentation, by contrast, can boost the skills of workers — helping radiologists make better diagnostic decisions, say, or enhancing fraud detection for banks. It can also create whole new classes of jobs. Robert Seamans, a professor of management and organizations at the New York University Stern School of Business, said some new roles are already in play: AI integrators that help companies deploy the technology,for example, or AI auditors who discern how and why the technology is doing things in certain ways. But it is hard to predict how entire occupations or even industries spring forth when new technologies are born. There were no electricians before electricity, no flight attendants before airplanes.

“It's almost a foregone conclusion there will be new work,” said Muro. The outstanding question for this particular wave of automation by AI, he added, was how much new work there might be.

A question of demand

Dietrich Vollrath, a professor of economics at the University of Houston, explains that there is a big open question around how the world responds to the commoditization of knowledge work that AI will provide. If AI makes knowledge work faster and cheaper, will the world want the same amount of knowledge work done? Or will it respond by demanding more and more complex knowledge work? 

Historical examples suggest that the trend could be one of increased demand. Making financial transactions cheap and easy by digitizing them helped integrate markets, partly automating the jobs of tellers, explained Lawrence Katz, a labor economist and professor of economics at Harvard University. But ultimately it also served to make the financial markets bigger, more interconnected and more complex, which meant more branches. The financial sector now employs more people than it did 40 years ago. “Every technological shock does threaten some jobs,” he said. But, he added, it has been “creative destruction.” 

A 2022 academic study co-authored by the MIT economist David Autor showed that, historically, technology has indeed done a pretty good job of creating new work: According to the study, 60 percent of employment in 2018 was in types of jobs that didn’t exist before 1940. But the study also found that, over the past 40 years, the impact of technologies that help automate tasks has intensified, while the impact of technologies that augment workers have not.

Much depends on the choices that are made. If more companies use AI to empower humans, there could be a flourishing of new capabilities and jobs; if more choose straight automation, we’ll see greater job loss. 

Fast or slow?

It is frustratingly difficult to predict the rate of adoption of AI, but “the faster these shocks occur, the more difficult the adjustment” will be, said Muro. He added that he is “somewhat skeptical about the rapidity of adoption” of AI across the workforce. He cited issues such as security and privacy of data as barriers that could slow down corporate adoption more than many people realize, especially in highly regulated sectors. Resistance from older or less tech-savvy managers could also slow things down, as could simple inertia, said Seamans.

Vollrath suggested that bottlenecks such as availability of electricity and specialized chips might also put a brake on AI’s spread by limiting supply of the technology that could raise the price of advanced systems. OpenAI has, for instance, reportedly been considering charging as much as $20,000 per month for access to its most advanced agents. “All of a sudden you can start to talk yourself into thinking it might just be cheaper to keep on Phil and Eddie and Gladys here in the HR department,” said Vollrath.

Yet other forces could speed things up. Another pandemic, a climate-related catastrophe or war, could push companies to automate faster, much as Covid-19 sparked a surge in remote work, argued Seamans.

Meanwhile, there are reasons for optimism about augmentation. Maria del Rio-Chanona, an assistant professor at University College London, argued that humanity faces massive unsolved challenges — climate change, medicine and more — that will need both human insight and AI support. “I don’t think we’re going to run out of problems [that need solving]” she said.

Forks in the road

It is impossible to predict exactly what all of this will look like as it plays out, but a rough outline is coming into view.

Vollrath pointed out that some business units that are highly exposed to automation, such as human resources or accounting, may simply wind down, shrinking gradually as new hires are frozen and attrition takes its course. Others might grow as demand for their services increases. And still other new departments, previously impossible to imagine, may bubble up. For now, it seems junior roles will be the most affected. If automation does reduce the number of junior workers that are required, it will also likely increase competition for remaining roles, and in doing so drive down salaries, said Seamans. 

“The lower end of the ladder of economic and social mobility may be weakened to the point that new young workers can't really get a start,” said Muro.

There may also be some false starts. Munyikwa pointed to the so-called productivity J-curve, a concept that predicts a dip in productivity as companies adopt new tools, before a big surge occurs. “We're kind of in that dip,” she said, adding that some companies might find that an uncomfortable place to be when AI projects don’t play out as hoped. That may help explain why Klarna, which launched an AI assistant to do the job of hundreds of customer service agents in 2024, quietly started hiring humans to work with the AI agents after customers complained about service. Muro predicts more missteps “made by firms that think they are in a position to lay people off and then find they can't.”

How governments and companies respond to all of this will shape the experience for millions. Universal basic income, stronger worker protections, retraining programs, or tax incentives for augmenting rather than replacing workers could all be on the table, according to experts who spoke with Aventine. For now, tech companies seem to be raising the alarm more than governments and other sectors, potentially because they’re seeing the effects of AI first. But “there are a lot of choices” that can be made to change how this affects people, said Katz. 

Along the way, there will be widespread change, discomfort for many, and a gradual emergence of new work.

“What did we see with the internet and computing?” asked Seamans. “It's not that we saw tons of jobs disappearing or for that matter, tons of new jobs being created. Instead, what we saw is most jobs changed, and in some cases, changed dramatically as a result of this technology. I think that that's a pretty good prediction for what's going to happen with AI.”

AI is transforming advertising, but maybe not how you expect. Artificial intelligence is, in many ways, an advertiser’s dream: It can spot patterns in vast datasets, target ads with precision and whip up synthetic images and videos to help lower production costs. In a dispatch from the advertising industry’s annual awards at the Cannes Lions International Festival of Creativity, The Economist confirms that these advances are reshaping the field. But the most unusual shift might be happening quietly in the background: As people turn to AI assistants for product recommendations instead of traditional search, it’s the AI models and the data they’re trained on that influence what people are buying, not the carefully crafted ads that have been targeted at consumers. Advertisers are already adapting. Some are now creating content specifically designed to be picked up by AI models, such as detailed product descriptions instead of flashy infographics, hoping to sway large language models into promoting their products. They’re also targeting online communities like Reddit and Facebook, knowing that these forums feed into important AI training sets used by the likes of OpenAI and Meta. The weirdest twist? The possibility that soon AI agents — not humans — could be authorized to make purchases themselves. That could bring about a strange new future, in which advertisers must figure out how to conduct business when both the creators and consumers of ads are machines.

Move over, GPS, better positioning systems are getting ready for prime time. Many of us would, if you’ll excuse the pun, be lost without the Global Positioning System, or GPS. Its presence in our phones is now essential for getting around, spying on our children and accessing ride-share apps. (It also facilitates far more consequential activities like syncing power grids, guiding offshore drilling and tracking shipping containers.) But despite its ubiquity, GPS has some significant shortcomings. For starters, it’s only accurate to within 15–30 feet, making it unreliable for precise tasks like keeping autonomous cars within a lane. And because GPS satellites orbit so high above Earth, their signals are weak at ground level, meaning GPS often fails indoors and is vulnerable to jamming and spoofing, as seen recently in conflict zones and around domestic airports. But solutions could be on the horizon, reports MIT Technology Review. Advancements in technology and better space launch economics have made building satellites that can operate closer to Earth more affordable. Xona Space Systems, a startup near San Francisco, plans to launch a constellation of 258 satellites over the next five years, placing them in low Earth orbit, about 12,000 miles closer to the planet than today’s GPS satellites. The intended result: Signals that could be up to 100 times stronger and positioning accuracy down to three inches. Xona’s system is designed to work alongside and complement existing GPS, improving coverage and providing a backup when necessary. The company launched its first satellite in late June; a handful more will be needed to begin augmenting the GPS network. 

A brain-computer interface can generate speech almost instantaneously. Researchers have enabled a man with a severe speech disorder due to ALS to synthesize speech that sounds like his own using an AI model — a breakthrough that offers new hope for people who have lost their ability to talk. The system, developed at the University of California Davis’s Neuroprosthetics Lab and tested on a 45-year-old participant, uses 256 electrodes implanted in the part of the brain responsible for speech movements. (By detecting only planned muscle activity, the researchers say, their setup avoids reading a user’s private inner thoughts by focusing only on intended speech.) In a paper published in Nature, the team describes collecting brain activity as the participant tried to speak sentences aloud. These signals were used to train an AI model to predict his intended speech. The AI model’s output was then fed to a  system built from recordings of his voice, made when he could still speak. The result conveys the inflections of natural speech, including questions, synthesized with just a 25-millisecond delay. In tests, human listeners understood the synthesized sentences 56 percent of the time, compared to just 3 percent of the time when they tried to understand the volunteer’s unaided speech. The technology is still experimental, but the researchers tell IEEE Spectrum that they plan to add more electrodes and improve the underlying AI models — steps that could eventually help many patients regain their voices.

Why the world cannot quit coal, from the Financial Times
2,900 words, or about 12 minutes

In terms of carbon emission per unit energy, coal is the most polluting energy source on the planet. But from an economic perspective, it remains irresistible: cheap, abundant, energy-dense and easy to stockpile and transport. This explains why it has always been a struggle for countries to stop using the fuel. While the Paris Agreement, signed almost a decade ago, looked like it might spell the end of coal, global use is now at an all-time high. Since 2000, global coal consumption has doubled, and the International Energy Agency has stopped predicting a peak and instead now foresees a plateauing of coal use around 2027. There are some notable examples of nations quitting coal entirely for the production of electricity, namely the UK, Austria and Portugal. But those gains are dwarfed by the coal use of India and China. India uses the fuel for 75 percent of its power generation and doesn’t plan to reach net zero emissions until 2070. And while coal’s share of China’s energy mix is shrinking, its absolute use continues to grow. The country is adding the equivalent of Canada’s entire electricity demand to its grid each year for the next three years, and its coal plant construction is now at a 10-year high. Also contributing to coal’s stubborn persistence in our energy mix are the destabilizing effects of the Covid-19 pandemic and the war in Ukraine, both of which pushed many countries to prioritize energy security over decarbonization. The bottom line: Even if the world does quit coal, it is going to take far longer than many people had hoped.

Brain Freeze, from Asterisk
5,700 words, or about 23 minutes

The dream of cheating death or glimpsing the distant future has long inspired interest in cryopreservation: the freezing of the body, or just the brain, in hopes of one day being revived. The idea has been around since the 1960s, but the seemingly simple notion of freezing a brain is actually surrounded by staggeringly complex science. For cryonics to have any hope of success, the brain must be preserved within as little as 12 minutes after death to prevent irreversible degradation. And even when that is achieved, scientists must still contend with the damage caused by the freezing process itself, which can shred delicate brain tissue. This story, written by a researcher in the field, traces how cryonics has gradually shifted from science fiction to experimental science. Progress is being made: New preservation techniques aim not just to limit tissue damage, but to preserve the brain’s information-containing structure, particularly the arrangement of synapses, which some believe hold the key to identity and memory. Still, practical cryonics is a long way off. It requires meticulous planning, highly controlled conditions at the moment of death and techniques that remain unproven. After more than 60 years of speculation, the field may only now be reaching a serious starting point. 

Futureproof, from The Verge
14,500 words, or about 44 minutes, across 12 stories

People working in creative industries are alarmed about what AI might mean for their livelihoods, with headlines dominated by fears of job losses and the use of their work for training data. Far less attention is paid to how these professionals are adapting. This package of stories explores how artists, musicians, writers, designers and even perfumers are experimenting with AI, planning for a future alongside it, and reimagining what it means to be creative as the landscape shifts. Among the more offbeat examples: AI-powered fonts that morph to suit different contexts, and the explosion of new fragrances as AI helps companies invent scents faster than ever. One standout story describes the development of artwork for a computer game made possible by generative AI, but brought to commercial success only after humans refined and enhanced it. It’s a fascinating portal into the current moment, in which AI can accelerate creative work but hasn’t yet replaced the need for human creativity, insight and skill.