GenAI Is Coming for Remote Workers First

There’s a tidy story out there when it comes to automation: If anyone felt its bite, it was workers in manufacturing and trade jobs first. The advent of the mechanized loom pushed weavers who worked for themselves into factories. After Ford trotted out the assembly line, skilled mechanics and engineers had to learn to go through rote motions. Eventually, support staff in offices experienced a similar fate. When networked PCs arrived, clerical and administrative workers saw their jobs shift from interacting with people to filling out software forms and checking boxes. All too many of them paid less. All too many build fewer skills. All too many treated with less variety, human connection, and dignity.

Meanwhile, most knowledge workers have seen roughly the opposite: automation has heralded more challenging, creative, skill-enhancing work that pays better — in dollars and status. And while it shocked us all, the pandemic and the new world of remote work gave these employees even more autonomy and a more balanced life. This isn’t just anecdotal; over five decades of research in fields such as labor economics, psychology, and sociology point to the fact that white-collar workers stand more to gain from automation. A senior manager with an MBA faces a lot of nonroutine, judgment-based tasks that produce a lot of value, and she benefits disproportionately from technology that automates routine work out of her way. In fact, scientists have a name for this “riches to the rich” phenomenon: skill-biased technical change.

Now, the story seems to be shifting.

Now it’s about physical presence. If you work entirely remotely — or could — many of your tasks may soon be heavily automated, while workers who have to show up and use their bodies for results are on safer ground. This means that some remote workers will lose their jobs, but many, many millions of them will soon experience massive job change, and therefore face a steep reskilling challenge.

To understand why, we need to start with new predictive models about generative AI “exposure.” According to recent research, the more exposed jobs are to GenAI, the bigger their potential productivity boost. For instance, Daniel Rock (a collaborator at Wharton) and his OpenAI coauthors found that if customer service reps picked up GenAI — just in its unrefined, hallucinating form — they could see 50% productivity gains in half of their tasks. Blockchain engineers, writers, and mathematicians? Totally exposed in this sense. Overall, the authors’ modeling suggests 80% of working adults have jobs that are 10% exposed, while 19% have jobs that are at least 50% exposed. Either way, most of us stand to gain a big productivity boost through GenAI use.

In fact, we’re already seeing this incentive at work. Microsoft’s recent workforce survey (of over 31,000 people across 31 countries) shows that as of May, 75% of employees surveyed are using GenAI in their daily work — a figure that has nearly doubled since January — and 78% of those workers reported that they adopted these tools independently. And their employers aren’t far behind: Gartner’s October 2023 survey indicated that 55% of corporations were either piloting or releasing LLM projects.

To most of us, the pace and scope of GenAI adoption are staggering. But from a historical and macroeconomic perspective, this could all spell a simply unusual rate of job change. GenAI was instantly available, free, and useful for 2.6 billion of us (80% of working adults around the globe — simple arithmetic implied by the GPTs paper linked above), so we will rapidly adjust how we work to accommodate, compared to prior general purpose technologies. Internal combustion engines, telephony, and the internet took multiple decades to diffuse; they were either costly, hard to distribute, or required new infrastructure. Those extra decades gave us an opportunity to adapt. We had accountants, barbers, and engineers before all of those technologies, and we have a lot more of them now, yet they do quite different things.

There are three larger forces at work that disturb that “disruption as usual” scenario. These should give us all pause and lead us to anticipate even more rapid and extensive job change to come — especially for remote workers.

Ease of use

The first is lying in plain sight: GenAI is being harnessed via easy-to-use applications. Think of GenAI as the arrival of electricity and the electric motor: an extremely powerful resource that could be used for lots of things in theory, but that isn’t all that great for any particular job right away. After electricity, we got dishwashers, water pumps, circular saws, and the like. Many firms are building the equivalent right now for generative AI. Some features of Microsoft’s Copilot like their “writing coach” — built into its Office suite — are a clear example. The aformentioned paper suggests that as this occurs, exposure (and thus utility) will dramatically increase for many workers: for instance, the average job in the U.S. will be 47–56% exposed, not 10%.

Organizations are already primed for GenAI

Second, firms have also preconfigured themselves to make GenAI automation easier: Since at least the advent of telephony, but especially in response to COVID, they have concentrated digital tasks to allow millions to work remotely. McKinsey’s 2020 COVID-19 digital transformation report summarizes this nicely: The average executive presumed a useful increase in “remote working or collaboration” would take 454 days, when it actually took 10.5. You read that right: 43 times faster than expected. And Upwork’s recent survey indicates that 22% of the U.S. workforce will be fully remote by 2025. GenAI can only handle digital tasks at the moment — those that deal in information inputs and outputs. This means that if you can work remotely all the time, GenAI can be profitably used to help with your tasks more than someone who has to show up in person.

GenAI is becoming more autonomous

The third force is an imminent uptick in GenAI autonomy. The GenAI we know is mostly passive. Prompt it and it does the work — but only a tiny bit of work, and with very poor, limited memory.

But OpenAI, Google, and Anthropic — the three titans in GenAI — have all been quite transparent: They expect to deliver highly autonomous systems that solve these problems in the next three to six months. In fact, for about four months now, we have had a public beta of one such agent: Devin. Unlike all the GenAI that came before, you can give Devin a goal — a pretty vague one at that — and it will come up with a plan to achieve that goal and get to work on it in the background. Via a simple chat interface, you can ask Devin how the work is coming along. It will text you back a progress update, like a colleague might via Slack. It will fail, but then recover. Repair. Improve. Innovate, even. Devin can often get the whole job done. ChatGPT couldn’t even get started here.

Soon, you will be able to delegate complex tasks to any major GenAI system and it will proactively handle large chunks of work — including asking questions for clarity and spinning up small teams of agents that they can delegate to. Any GenAI user will effectively have to learn how to manage automated organizations of artificial agents.

Some — maybe more than we’re used to — will lose their jobs. A serious issue for the tens of thousands affected that requires proactive attention from organizations and regulators. But for perhaps a hundred million remote workers around the globe, these forces will conspire to facilitate relatively dramatic, rapid job change.

The proper focus in response to this inbound tsunami of change is reskilling. Firms and workers must invest in the learning and development required to prepare workers and help them adjust as these changes unfold. To pick just one clear target: If you’re a remote worker, you should focus on building management skills. Whatever you did before, now you should probably learn to supervise a band of highly autonomous software agents to do it for you. This includes skills like delegation, which means clearly specifying a job you want done, ensuring you have clarity before work begins, and giving clear feedback to guide performance, just like when managing other people.

You might presume you don’t have to be concerned about motivating agents and tending to their career path, but you’d only be (for now) half right. This will require AI-specific skills. It turns out that agents respond quite differently to the social aspects of our prompts, and “motivating” them properly can make a big difference. But more broadly, firms that fail to address this reskilling challenge will not only miss the true productivity upside of GenAI; by failing to prepare them, they may also alienate the very human workforce that — at least for the foreseeable future — remains at the heart of their competitive advantage.

Matt Beane is an assistant professor of technology management at the University of California, Santa Barbara, and a research affiliate with MIT’s Initiative on the Digital Economy. He is the author of The Skill Code: How to Save Human Ability in an Age of Intelligent Machines.

This article is adapted from Harvard Business Review with permission. ©2024. All rights reserved.