Jobs, tasks and the economics of automation
Robots aren't yet ready to clean our homes or serve us coffee - or at least, they can do so only in very clearly defined way. However, AI is becoming increasingly capable of cognitive tasks like writing, analysis, and customer service. Much of this work can be done remotely, which means it's possible to automate it now or in the near future.
That raises many questions; recent research from Epoch AI tackles one of the biggest: what would happen to our economy if AI could fully automate all remote work?
Understanding remote work's role
First, we need to understand how much of our work can actually be done remotely. During the COVID-19 pandemic, many of us experienced firsthand the shift to remote work. Previous research suggested that about 37% of all U.S. jobs could be performed entirely from home. However, for this analysis, Epoch AI has looked at specific work tasks rather than entire jobs.
The research suggests that about 34% of all work tasks can be performed remotely, that is, using just a computer and internet connection. This is a crucial distinction because automation rarely makes entire jobs obsolete overnight. Instead, it typically starts by automating specific tasks within jobs. I already see this effect in many companies where specific workflows have been automated, from procurement to screening job applicants.
The substitution effect
An essential concept in the research is what economists call elasticity of substitution. What they mean is simply a measure of how easily one type of work can replace another.
For example, when the pandemic forced many people to work from home, there were two scenarios:
If remote and in-person work couldn't easily substitute for each other, we would have seen a massive drop in economic output.
If they could substitute well, the economy would adapt more smoothly.
What actually happened? The economic impact was surprisingly modest. U.S. GDP fell by about 9.2% initially – a significant drop, but far less than many economists predicted given that roughly a third of workers suddenly shifted to remote work. This suggests that remote work can substitute for in-person work more effectively than previously thought. Substitution is remarkably elastic.
With this in mind, the researchers modeled two scenarios:
Conservative Scenario: Assuming remote and non-remote work don't substitute well for each other.
U.S. GDP could still double. This would mean 7% annual growth if spread over a decade
Optimistic Scenario: Assuming high substitutability (which the pandemic data suggests is more likely)
The economy could grow by a factor of 10 or more. This would mean over 25% annual growth if spread over a decade
That's staggering. To put these numbers in perspective, the U.S. economy typically grows by about 2-3% annually. Even the conservative scenario represents a dramatic acceleration of economic growth.
Is this even possible?
How could such dramatic growth occur? The study suggests it would happen by deploying digital workers – AI systems that can perform remote tasks. These could be hosted in data centers, potentially billions of digital workers operating simultaneously.
This is what Jensen Huang of NVidia was talking about CES when he said:
In a lot of ways, the IT department of every company is going to be the HR department of AI agents in the future. Today, they manage and maintain a bunch of software from the IT industry; in the future, they will maintain, nurture, onboard, and improve a whole bunch of digital agents and provision them to the companies to use.
For businesses, this could mean dramatically increased economic productivity without corresponding increases in labor costs, but even more likely, new business models built around AI-powered services.
Good work and the human impact
This is all dramatic enough, but we must consider the broader implications for their workforce and society.
As I said, history shows that automation tends to transform jobs rather than eliminate them entirely. So, as AI takes over routine remote tasks, workers may move into new roles that complement automated systems.
So we could focus on identifying which tasks, not jobs, can be automated and create new roles that leverage human creativity and judgment to complement those tasks with even greater value.
As automation increases, we'll likely see growing demand for certain skills involving higher-level cognitive abilities. For example, technical literacy, AI literacy, strategic thinking, creative problem-solving, emotional intelligence and complex communication will all be essential qualities in the new market.
These skills may become more necessary, but we still need to hope we can create not only more work but Good Work.
Agents and Agency
The principles of Good Work offer a framework for managing the transition in ways that benefit both the business and its workforce. The basis is an understanding that economic security must evolve alongside technological change. This is not simply preserving jobs; it may well require us to reimagine how work is structured and compensated: new compensation frameworks that reflect the changing nature of work in hybrid human-AI environments.
While AI agents may make the headlines just now, the agency of human workers will be critical to the well-being of the workforce and society as a whole. By agency, I mean giving employees meaningful control over their interactions with AI systems rather than making them passive subjects of AI priorities.
Perhaps most critically, successful automation requires genuine worker participation throughout the process. Organizations that excel in this transition will actively involve their workforce in automation planning and implementation. This approach should lead to better technical outcomes but also help build trust and buy-in across the organization.
Given the economic potential of AI, I have no doubt these changes will rush up on us over the next few years. The easy part will be the automation and intelligence. The hard work will be creating the Good Work.