Maya Kulycky loves a bulleted list, but even more, she loves handing over the drudgery to a digital assistant. “I really like the drafting tool,” she says of Gemini in Google Docs, the AI function embedded across several apps in Google Workspace. She starts with a rough draft of a memo—“Here’s what we had a conversation around, here’s what we want to do, here are the next steps we’re going to take, here’s the timeline”—and then asks Gemini to arrange those thoughts into a well-formatted email to her team at Google Research. “The technology is with me,” she says. “It’s not a replacement for me. It’s an accelerant.”
As AI arrives at our workplaces, it’s showing up as a helpful tool, not as a superintelligence. The signs are clear that AI will affect every sector of the workforce, from entry-level employees to managers and CEOs. The deployment of AI is both a design and a user-interface problem and a wider, societal responsibility of technologists, AI companies, and government and legislators. The next few years are critical. We need to lay a proper foundation for AI at work.
In her role as vice president for strategy, operations, and outreach, Kulycky is tasked with aligning the teams that push significant AI innovations forward. Kulycky believes we need to proceed carefully to ensure that AI augments human potential. One of the near-term goals is exploring how AI can assist with the many background tasks and organizational work that often bog down teams, the so-called day-to-day that can get in the way. If these commitments could be offloaded onto AI agents, workers would have more time for the most important and creative challenges they face.
The future that AI creates will also depend upon how exactly we integrate it. Take the example of introducing customer-service agents that are entirely AI-based. A company can choose to provide better, more thoughtful care to customers (with the AI assisting humans), or the company could cut jobs. The economist Erik Brynjolfsson studies how the workforce will be affected by AI. When asked about the specter of workers being replaced by AI, Brynjolfsson brought up the Luddites, the famous reactionary group. Concerned that looms and automatic weaving machines would replace skilled artisans, knitters broke into factories to tear down the machinery that was automating away their jobs. But although their protest made history, it didn’t halt technological progress. “A lot of specific jobs did disappear,” Brynjolfsson says. By the late 1700s, the Industrial Revolution was increasingly disrupting the textile industry. “But it led to new jobs for other people—and economic growth and higher living standards.”
The pattern has repeated itself through a series of technological developments. In the United States, major technological shifts such as automobiles and computerization have not led to long-term widespread unemployment. “We’ve had lots of new technology, but there are still a lot of jobs. It’s just that there are different jobs to do,” Brynjolfsson says. For example, from 1900 to 2020, the proportion of Americans working in agriculture dropped from 42 percent to 2 percent. “Most of us are doing other things,” Brynjolfsson says, “and the jobs that are normal to us now are jobs that people couldn’t have even imagined before.” A 2022 research paper by the economist David Autor found that 60 percent of working Americans currently occupy roles that didn’t exist in the 1940s. We’ve shifted from mining to computer coding, from textile manufacturing to health care.
Each week, it seems, we receive news of bigger and more powerful large language models (LLMs), and that leads the public imagination to a place where AI tools are so sophisticated that they severely pressure huge segments of the workforce. Brynjolfsson offered a note of optimism. While the capabilities of machine learning are certainly remarkable, he cautioned against presuming that machines will fully automate modern employment processes. “That does work sometimes. But more often, you want to keep a human in the loop because the technology is not all-encompassing,” he says. “When we break down jobs into tasks, we find that AI can often do a core set of tasks well. But humans are better at dealing with exceptions, improvising, dealing with things that come up.”
According to a working paper from the International Monetary Fund (IMF), AI will likely enhance rather than replace jobs that have a lot of in-person interaction, critical decision-making, and specialized expertise. Occupations such as lawyers, surgeons, or judges will benefit both from AI’s ability to take on routine administrative tasks and from its assistance in core functions such as analyzing medical scans or writing a first draft of a patent submission. Other studies have shown that AI can help people become faster and better at tasks such as business writing, programming, and consulting.
Brynjolfsson and his colleagues ran a now-famous study for which they observed the implementation of an LLM in a call center. The AI examined the calls completed by the most effective employees, then surfaced those practices to everyone else in the work group. The least-skilled workers benefited the most from AI; their performance jumped by 35 percent. AI’s ability to close the gap between high and low performers could also streamline onboarding, acclimating workers quickly to new roles. That benefit could be a double-edged sword: It’s great for helping employees succeed quickly on the job, but at the organizational level, it lowers the barrier to replacing workers, which in turn could erode job security.
Service professions will also benefit from assistive AI, according to the IMF study. Customer service chatbots can handle basic requests, freeing agents to contend with more complex cases. For example, Best Buy is resolving customer issues up to 90 seconds faster using Google Cloud’s automated call summarization. A research paper by Barclays calls this shift “service-ization”—a nod to industrialization, a process by which the hands-on, intensive work in manufacturing and agriculture became the work of machines.
Though the research by scholars and analysts points to the productivity gains that can result from the assistance of AI, businesses’ uptake has been slow. Brynjolfsson expressed surprise about how the adoption of artificial intelligence is already lagging behind the technology’s capabilities: “I’m disappointed by how few people have thought about how to use generative AI (GenAI) in their daily work. While the technology is racing ahead very rapidly, there’s much less energy put into how we get business value out of it.” Part of the reason for that lag could be the investment it takes to implement workflows around the adoption of any new tool. Still, Brynjolfsson encourages leaders to consider how these new technologies can improve their businesses. “In 2017, I wrote that AI won’t replace managers, but managers who use it will replace managers who don’t. That’s still true today.”
Even at the current rate of adoption, experts predict that AI will eclipse certain sectors of labor while creating opportunities for new kinds of jobs. Goldman Sachs estimates that GenAI may automate close to a quarter of all jobs. A 2023 McKinsey report predicts that 30 percent of today’s work hours could be replaced by GenAI as soon as 2030. That same report found that low-wage workers (those earning less than $38,200 a year) are up to 14 times more vulnerable to job elimination than America’s highest earners. Brynjolfsson agrees: In a 2017 paper, he found that machine learning will most profoundly affect low-wage workers.
That poses both economic and social concern since low-wage jobs are disproportionately held by women, people of color, and workers who haven’t received higher education. Anyone affected by such an industry shift will have the added burden of needing to learn new skills to transition to a new role successfully. AI can assist with that training, given how quickly it can do the work of, say, personalizing learning programs and translating content.
Perhaps the simplest concern about the economic impact of AI at work is who will benefit financially from these large shifts. Disruptive technologies—even those that eliminate jobs—historically raise national income because they boost overall productivity. But the workers affected by that shift don’t necessarily see that economic benefit. While those who transition successfully may continue to collect the labor wages they rely on, the benefit of their increased productivity tends to be concentrated among corporate leadership. “New technology has the possibility to close the skills gap between workers, but it can increase the gap between capital and labor,” says Brynjolffson. “The best strategy is to aim for technology that creates widely shared prosperity, not just concentrated wealth.”
Companies will need to work together, along with governments and nonprofits, to help more people realize the economic potential of AI. One example of such partnership is the AI-Enabled ICT Workforce Consortium, founded by Cisco in partnership with eight other companies including Accenture, Eightfold, Google, IBM, Indeed, Intel, Microsoft and SAP, which will recommend skilling and upskilling opportunities to ensure workers can adapt. Six members of this consortium have committed to goals for training, upskilling, and reskilling millions of people over the next 10 years; Google’s $120 million Global AI Opportunity Fund helps workers and students from all backgrounds access AI training —including the company’s AI Essentials course—in local communities at no cost, building on the more than 100 million people Google has trained in digital skills to date.
In places like parts of Europe and Asia where the working-age population is falling, AI can complement the workforce, taking on tasks that allow workers to focus on the most important, rewarding aspects of jobs. The other good news, per Brynjolfsson: “In that world, productivity would be dramatically higher—we’d have 2, 5, 10x more productivity per capita. Then, we’d be able to provide basic needs for people. As Keynes put it, the economic problem would be solved. People might or might not still enjoy working, but they wouldn’t need to worry about starving. A world like that is not imminent, but I can imagine the benefits of thinking about how we would run that economy.”
In the meantime, who is responsible for rallying industry around an economy-preserving, assistive approach to AI? Brynjolfsson recommends that entrepreneurs and executives focus on the top line as well as the bottom, broadening their performance goals to look not just at costs but also at metrics such as quality and customer satisfaction. He also imagines that organizations will get more buy-in from their customers and their workforce if they use AI to complement workers. “Many people would want to speed adoption of a tech that will help them do their job better, instead of one pitched as a replacement for them,” he says.
Countries can also choose to apply AI in their own ways. Some may use tools to combat misinformation, for instance, while others might lean on AI to create the world’s most effective censorship system.
Technologists, of course, also have the power to shape the industry. The benchmarks used to measure the effectiveness of machine learning models have been inspired by the Turing test, a test that measures the strength of a machine by how closely it can imitate human conversation. Establishing new kinds of benchmarks could help position AI as assistive—a paradigm shift that could lead to more impressive innovations. “Merely mimicking humans leads us into what I call the Turing trap. Instead, look at technologies that augment us to do things that no human could do before. They’ll create more value and shared prosperity,” says Brynjolfsson.
Introducing AI responsibly will take some creative thinking on the part of corporate leaders and government officials. Rather than clinging to existing job distribution, Brynjolfsson advocates offering tools for training and job matching. According to a 2024 survey from Deloitte, only 47 percent of leaders agree that their organizations are sufficiently educating employees on the capabilities, benefits, and value of GenAI. In addition to upskilling employees, businesses will need to change what qualifications they look for when hiring (it seems adaptability will become a crucial skill), opening their doors to candidates who may have less prior training or education but can quickly adjust and learn new skills. Going forward, success in the workplace will be less about possessing specific skills and more about being flexible, Brynjolfsson predicts. “The era of going to school for 12, 16, or more years, learning stuff, and using that for the next 40 years is gone,” he says. “People have to continually learn and adjust.”
